JP2024513995A - Multichannel protein voxelization to predict variant pathogenicity using deep convolutional neural networks - Google Patents

Abstract

The system includes at least a voxelizer, an alternative allele encoder, an evolutionary conservation encoder, and a convolutional neural network. The voxelizer accesses a three-dimensional structure of a reference amino acid sequence of the protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid. The alternative allele encoder encodes an alternative allele amino acid for each voxel in the three-dimensional grid of voxels. The evolutionary conservation encoder encodes an evolutionary conserved sequence for each voxel in the three-dimensional grid of voxels. The convolutional neural network applies a three-dimensional convolution to a tensor including the alternative allele sequences and the distance channels for each amino acid encoded with the respective evolutionary conserved sequences, and determines pathogenicity of the variant nucleotide based at least in part on the tensor.

Description

(Priority application)
This application is based on “Multi-channel Protein Voxelization To Predict Variant Pathogenicity Using Deep Convolutional Neural Netw” filed on March 24, 2022. No. 17/703,935 entitled ``Orks'' (Attorney Docket No. ILLM 1047-2/IP-2142-US), which was filed on April 15, 2021. ep Convolutional Neural Networks” Claims priority to U.S. Provisional Patent Application No. 63/175,495, entitled Attorney Docket No. ILLM 1047-1/IP-2142-PRV.

This application is also filed in U.S. Nonprovisional Patent Application No. 17/703,958, entitled "Efficient Voxelization For Deep Learning," filed on March 24, 2022 (Attorney Docket No. ILLM 1048-2/IP-2143- U.S. Provisional Patent Application No. 63/175,767 (Attorney Docket No. ILLM 1048-1/IP-2143-PRV).

The priority application is incorporated herein by reference for all purposes.

(Related application)
This application relates to a concurrently filed PCT patent application entitled "Efficient Voxelization For Deep Learning" (Attorney Docket No. ILLM 1048-3/IP-2143-PCT). Related applications are incorporated herein by reference for all purposes.

(Field of invention)
The disclosed technology is subject to uncertainties regarding artificially intelligent computers and digital data processing systems, and corresponding data processing methods and products for mimicking intelligence (i.e., knowledge-based systems, reasoning systems, and knowledge acquisition systems). systems for logical reasoning (e.g., fuzzy logic systems), adaptive systems, machine learning systems, and artificial neural networks. In particular, the disclosed techniques relate to using deep convolutional neural networks to analyze multi-channel voxelized data.

(built-in)
The following is incorporated by reference for all purposes as if fully set forth herein.
Sundaram, L. et al. Predicting the clinical impact of human mutations with deep neural networks. Nat. Genet. 50, 1161-1170 (2018),
Jaganathan, K. et al. Predicting splicing from primary sequence with deep learning. Cell 176, 535-548 (2019),
U.S. Provisional Patent Application No. 62/573,144 entitled “TRAINING A DEEP PATHOGENICITY CLASSIFIER USING LARGE-SCALE BENIGN TRAINING DATA” filed on October 16, 2017 (Attorney Docket No. ILLM 1000-1/IP-1611 -PRV),
“PATHOGENICITY” filed on October 16, 2017
CLASSIFIER BASED ON DEEP CONVOLUTIONAL NEURAL NETWORKS (CNNs)'' (Attorney Docket No. ILLM 1000-2/IP-1612-PRV);
U.S. Provisional Patent Application No. 62/573,153 (Attorney Docket No. ILL M 1000-3/IP- 1613-PRV),
U.S. Provisional Patent Application No. 62/582, 89 entitled “PATHOGENICITY CLASSIFICATION OF GENOMIC DATA USING DEEP CONVOLUTIONAL NEURAL NETWORKS (CNNs)” filed on November 7, 2017 No. 8 (Agent reference number ILLM 1000-4/IP- 1618-PRV),
U.S. Nonprovisional Patent Application No. 16/160,903 entitled “DEEP LEARNING-BASED TECHNIQUES FOR TRAINING DEEP CONVOLUTIONAL NEURAL NETWORKS” filed on October 15, 2018 (Attorney Docket No. ILLM 1000-5/IP-1611 -US),
U.S. Nonprovisional Patent Application No. 16/160,986 entitled “DEEP CONVOLUTIONAL NEURAL NETWORKS FOR VARIANT CLASSIFICATION” filed on October 15, 2018 (Attorney Docket No. ILLM 1000-6/IP-1612-US ),
U.S. Nonprovisional Patent Application No. 16/160,968 entitled “SEMI-SUPERVISED LEARNING FOR TRAINING AN ENSEMBLE OF DEEP CONVOLUTIONAL NEURAL NETWORKS” filed on October 15, 2018 (Agent reference number ILLM 1000-7/IP -1613-US), and U.S. Nonprovisional Patent Application No. 16/407 entitled “DEEP LEARNING-BASED TECHNIQUES FOR PRE-TRAINING DEEP CONVOLUTIONAL NEURAL NETWORKS” filed May 8, 2019. , No. 149 (Representative Arrangement No. ILLM 1010-1/IP-1734-US).

The subject matter discussed in this section should not be assumed to be prior art merely as a result of mention in this section. Similarly, it should not be assumed that the problems mentioned in this section or associated with the subject matter provided as background have been previously recognized in the prior art. The subject matter in this section merely represents different approaches, and as such may also correspond to implementations of the claimed technology.

Genomics in the broad sense, also called functional genomics, aims to characterize the function of all genomic elements of an organism by using genome-scale assays such as genome sequencing, transcriptome profiling and proteomics. Genomics has emerged as a data-driven science, operating not by testing preconceived models and hypotheses, but by discovering novel properties from the examination of genome-scale data. Applications of genomics include finding associations between genotype and phenotype, discovering biomarkers for patient stratification, predicting the function of genes, and biochemical studies such as transcriptional enhancers. This includes charting the genomic regions that are actively active.

Genomics data is too large and complex to be mined solely by visual inspection of pairwise correlations. Instead, analytical tools are needed to support the discovery of unexpected relationships, derive new hypotheses and models, and make predictions. Unlike some algorithms where assumptions and domain expertise are hard-coded, machine learning algorithms are designed to automatically detect patterns in data. Machine learning algorithms are therefore well-suited to data-driven science, especially genomics. However, the performance of machine learning algorithms can strongly depend on how the data is represented, that is, how each variable (also called a feature) is calculated. For example, to classify a tumor as malignant or benign from a fluorescence microscopy image, a preprocessing algorithm can detect cells, identify cell types, and generate a list of cell numbers for each cell type.

Machine learning models can take estimated cell count, an example of a manually designed feature, as an input feature for classifying tumors. The central problem is that classification performance is highly dependent on the quality and relevance of these features. For example, relevant visual features such as cell morphology, distance between cells or localization within an organ are not captured in cell counting, and this incomplete representation of the data can reduce classification accuracy.

Deep learning, a subdiscipline of machine learning, addresses this problem by embedding feature computation into the machine learning model itself, producing an end-to-end model. This result is achieved through the development of deep neural networks, machine learning models that involve successive elementary actions that compute increasingly complex features by taking as input the results of previous actions. Deep neural networks can improve prediction accuracy by discovering relevant features with high complexity, such as cell morphology and spatial organization of cells in the example above. The construction and training of deep neural networks has been made possible by the data explosion, advances in algorithms, and vast increases in computing power, particularly through the use of graphical processing units (GPUs).

The goal of supervised learning is to obtain a model that takes features as input and returns predictions of so-called target variables. An example of a supervised learning problem is the problem of predicting whether or not an intron will be spliced (target) by considering RNA characteristics such as the presence or absence of a standard splice site sequence, the position of a splicing branch point, or the length of an intron. be. Training a machine learning model refers to learning its parameters, which generally involves minimizing a loss function on training data with the purpose of making accurate predictions on unknown data.

For many supervised learning problems in computational biology, input data can be represented as a table with multiple columns or features, each of which contains numerical or categorical data that is potentially useful for making predictions. Contains. While some input data is naturally represented as tabular features (e.g. temperature or time), other input data can be adapted to fit into a tabular representation using a process called feature extraction. first needs to be converted (eg, converting deoxyribonucleic acid (DNA) sequences to Umer counts). For intron-splicing prediction problems, the presence or absence of canonical splice site sequences, location of splicing branch points, and intron length can be preprocessed features collected in tabular form. Tabular data is the standard for a wide range of supervised machine learning models, ranging from simple linear models such as logistic regression to more flexible nonlinear models such as neural networks and many others.

Logistic regression is a binary classifier, a supervised learning model that predicts binary target variables. Specifically, logistic regression predicts the probability of a positive class by calculating a weighted sum of input features mapped to the [0,1] interval using a sigmoid function, which is a type of activation function. . The parameters of logistic regression, or other linear classifiers that use different activation functions, are the weights in the weighted sum. A linear classifier fails, for example, if the class of whether an intron is spliced or not cannot be sufficiently distinguished by a weighted sum of input features. To improve prediction performance, new input features can be added manually by transforming or combining existing features in new ways, for example by taking powers or pairwise products.

Neural networks automatically learn these nonlinear feature transformations using hidden layers. Each hidden layer can be thought of as multiple linear models whose outputs are transformed by a nonlinear activation function, such as a sigmoid function or the more general rectified linear unit (ReLU). At the same time, these layers organize the input features into related complex patterns, facilitating the task of distinguishing between the two classes.

Deep neural networks use many hidden layers, and when each neuron receives input from all neurons in the previous layer, the layers are said to be fully connected. Neural networks are generally trained using stochastic gradient descent, an algorithm suitable for learning models on very large datasets. Neural network implementations using modern deep learning frameworks enable rapid prototyping with different architectures and datasets. Fully connected neural networks can predict the proportion of exons spliced for a given sequence from sequence features such as the presence of splice factor binding motifs or sequence conservation, and identify potentially disease-causing genetic variants. It can be used for several genomics applications, including prioritizing and predicting cis-regulatory elements in a given genomic region using features such as chromatin marks, gene expression, and evolutionary conservation. .

For effective predictions, local dependencies in spatial and longitudinal data must be taken into account. For example, shuffling DNA sequences or pixels of an image severely disrupts information patterns. These local dependencies set spatial or longitudinal data apart from tabular data where the ordering of features is arbitrary. Consider the problem of classifying genomic regions as bound versus unbound by a particular transcription factor, where binding regions are defined as high-confidence binding events in chromatin immunoprecipitations following sequencing (ChIP-seq) data. Transcription factors bind to DNA by recognizing sequence motifs. Fully connected layers based on sequence-derived features such as the number of k-mer instances in the sequence or position weight matrix (PWM) matches can be used for this task. Because k-mer or PWM instance frequencies are robust to shifting motifs within a sequence, such models can generalize well to sequences with the same motif located at different positions. . However, they are unable to recognize patterns in which transcription factor binding depends on a combination of multiple motifs with well-defined spacing. Furthermore, the number of possible k-mers increases exponentially with k-mer length, which poses both conservation and overfitting challenges.

A convolutional layer is a special form of a fully connected layer in which the same fully connected layer is applied locally, for example within a 6 bp window, to all sequence positions. This approach can also be viewed as scanning the sequence using multiple PWMs, for example for transcription factors GATA1 and TAL1. By using the same model parameters across locations, the total number of parameters is dramatically reduced and the network is able to detect motifs at locations not seen during training. Each convolutional layer scans the array with a number of filters by generating scalar values at every position that quantize the correspondence between the filter and the array. As in a fully connected neural network, a non-linear activation function (generally ReLU) is applied at each layer. A pooling operation is then applied, which aggregates the activations in consecutive bins across the position axis, generally taking the maximum or average activation for each channel. Pooling reduces the effective array length and coarsens the signal. Subsequent convolutional layers constitute the output of the previous layer and can detect whether the GATA1 and TALI motifs were present in a certain distance range. Finally, the output of the convolutional layer can be used as input to a fully connected neural network to perform the final prediction task. Thus, different types of neural network layers (eg, fully connected layers and convolutional layers) can be combined within a single neural network.

Convolutional neural networks (CNNs) can predict various molecular phenotypes based solely on DNA sequences. Applications include classification of transcription factor binding sites and prediction of molecular phenotypes such as chromatin features, DNA contact maps, DNA methylation, gene expression, translation efficiency, RBP binding, and microRNA (miRNA) targets. In addition to predicting molecular phenotypes from sequences, convolutional neural networks can be applied to more technical tasks traditionally addressed by hand-designed bioinformatics pipelines. For example, convolutional neural networks can predict guide RNA specificity, denoise ChIP-seq, improve Hi-C data resolution, predict laboratory origin from DNA sequences, and call genetic variants. can. Convolutional neural networks have also been used to model long-term dependencies in genomes. Although interacting regulatory elements may be located remotely on the unfolded linear DNA sequence, these elements are often proximal in the actual 3D chromatin conformation. Therefore, modeling molecular phenotypes from linear DNA sequences is a crude approximation of chromatin, but allows for long-term dependence and allows the model to implicitly learn aspects of 3D organization such as promoter-enhancer loops. This can be improved by making it possible. This is achieved by using dilated convolution with a receptive field of up to 32 kb. Extended convolution also allows splice sites to be predicted from the sequence using a 10 kb receptive field, thereby allowing integration of gene sequences over a distance as long as a typical human intron ( Jaganathan, K. et al. Predicting splicing from primary sequence with deep learning. Cell 176, 535-548 (2019)).

Different types of neural networks can be characterized by their parameter sharing schemes. For example, fully connected layers have no parameter sharing, whereas convolutional layers impose translational invariance by applying the same filter at every position of their input. Recurrent neural networks (RNNs) are an alternative to convolutional neural networks for processing sequential data such as DNA sequences or time series, implementing different parameter sharing schemes. Recurrent neural networks apply the same operation to each array element. This operation takes as input the memory of the previous array element and a new input. It updates memory and optionally emits output, which is either passed to subsequent layers or used directly as model predictions. By applying the same model at each array element, the recurrent neural network is invariant to position index in the processed array. For example, recurrent neural networks can detect open reading frames in a DNA sequence regardless of their position in the sequence. This task requires recognition of a specific set of inputs, such as a start codon followed by an in-frame stop codon.

The main advantage of recurrent neural networks over convolutional neural networks is that, in theory, they can pass information through infinitely long arrays through memory. Furthermore, recurrent neural networks can naturally handle sequences of widely varying lengths, such as mRNA sequences. However, convolutional neural networks combined with various tricks (such as dilated convolution) reach comparable or even better performance than recurrent neural networks for sequence modeling tasks such as audio synthesis and machine translation. be able to. Recurrent neural networks can aggregate the output of convolutional neural networks to predict single cell DNA methylation status, RBP binding, transcription factor binding, and DNA accessibility. Furthermore, since recurrent neural networks apply sequential operations, they cannot be easily parallelized and are therefore much slower to compute than convolutional neural networks.

Although most of the human genetic code is common to all humans, each human has a unique genetic code. In some cases, the human genetic code may contain outliers, called genetic variants, that may be common among a relatively small group of individuals in the human population. For example, a particular human protein may contain a particular sequence of amino acids, whereas variants of that protein may differ by one amino acid in an otherwise identical particular sequence.

Genetic variants can be pathogenic and cause disease. Most such genetic variants have been depleted from the genome by natural selection, but the ability to identify which genetic variants are likely to be pathogenic is limited by the ability of researchers to identify these genetic variants. The focus can help to gain an understanding of the corresponding diseases and their diagnosis, treatment, or cure. The clinical interpretation of millions of human gene variants remains unclear. Some of the most frequent pathogenic variants are single nucleotide missense mutations that change amino acids in proteins. However, not all missense mutations are pathogenic.

Models that can directly predict molecular phenotypes from biological sequences can be used as in silico perturbation tools to investigate the association between genetic and phenotypic variation, and can be used to quantitatively predict trait loci. It has emerged as a new method for identification and variant prioritization. These approaches recognize that the majority of variants identified by genome-wide association studies of complex phenotypes are non-coding, which makes it difficult to estimate their effects and contributions to the phenotype. It is very important to consider. Furthermore, linkage disequilibrium results in blocks of co-inherited variants, which makes it difficult to pinpoint individual causative variants. Therefore, sequence-based deep learning models that can be used as matching tools to assess the impact of such variants provide a promising approach to finding potential drivers of complex phenotypes. As an example, differences between two variants can indirectly predict the effect of non-coding single nucleotide variants and short insertions or deletions (indels) on transcription factor binding, chromatin accessibility or gene expression prediction. Can be mentioned. Another example is predicting novel splice site generation from the quantitative effects of genetic variants on sequence or splicing.

An end-to-end deep learning approach for variant effect prediction is applied to predict the pathogenicity of missense variants from protein sequences and sequence conservation data (Sundaram, L. et al. Predicting the clinical impact of human mutation with deep neural networks. Nat. Genet. 50, 1161-1170 (2018), herein referred to as “PrimateAI”). PrimateAI uses deep neural networks trained on known pathogenic variants with data augmentation using cross-species information. In particular, PrimateAI uses wild-type and mutant protein sequences to compare differences and uses trained deep neural networks to determine the pathogenicity of mutations. Such approaches that utilize protein sequences for pathogenicity prediction are promising as they can avoid roundness problems and overfitting to prior knowledge. However, the amount of clinical data available in ClinVar is relatively small compared to the sufficient amount of data to effectively train deep neural networks. To overcome this data gap, PrimateAI uses common human variants and primate-derived variants as benign data and simulated variants based on trinucleotide context as unlabeled data. did.

PrimateAI outperforms traditional methods when trained directly on sequence alignments. PrimateAI learns important protein domains, conserved amino acid positions, and sequence dependencies directly from training data consisting of approximately 120,000 human samples. PrimateAI substantially outperforms other variant pathogenicity prediction tools in distinguishing between benign and pathogenic de novo mutations in candidate developmental disorder genes and in reproducing prior knowledge in ClinVar. These results suggest that PrimateAI is an important advance for variant classification tools that can reduce reliance on prior knowledge of clinical reporting.

Central to protein biology is the understanding of how structural elements give rise to observed functions. The plethora of protein structural data allows the development of computational methods to systematically derive rules governing structure-function relationships. However, the performance of these methods depends critically on the choice of protein structure representation.

Protein sites are microenvironments within protein structures that are distinguished by their structural or functional roles. A site can be defined by a three-dimensional (3D) location and a local neighborhood around this location in which the structure or function resides. Central to rational protein engineering is an understanding of how the structural arrangement of amino acids creates functional features within protein sites. Determination of the structural and functional roles of individual amino acids in proteins provides information to aid in the manipulation and modification of protein function. Identification of functionally or structurally important amino acids enables focused engineering efforts such as site-directed mutagenesis to modify the functional properties of a target protein. Alternatively, this knowledge can help avoid engineering designs that disable desired functionality.

Since it is established that structure is much more conserved than sequence, the increase in protein structural data provides an opportunity to systematically study the underlying patterns governing structure-function relationships using data-driven approaches. provide. A fundamental aspect of any computational protein analysis is how protein structural information is represented. The performance of machine learning methods often depends more on the choice of data representation than on the machine learning algorithm used. A good representation efficiently captures the most important information, whereas a bad representation produces a noisy distribution with no underlying pattern.

The recent success of a plethora of protein structures and deep learning algorithms provides an opportunity to develop tools to automatically extract task-specific representations of protein structures. Therefore, an opportunity arises to predict variant pathogenicity using multichannel voxelized representations of 3D protein structures as inputs to deep neural networks.

The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

Color drawings may also be available in patent application information retrieval (PAIR) via the Supplemental Content tab. In the drawings, like reference characters generally refer to similar parts throughout the different figures. Additionally, the drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the disclosed technology. In the following description, various implementations of the disclosed technology are described with reference to the following drawings.

1 is a flowchart illustrating the process of a system for determining pathogenicity of a variant, according to various embodiments of the disclosed technology. FIG. 1 schematically depicts an exemplary reference amino acid sequence of a protein and an alternative amino acid sequence of a protein, according to one embodiment of the disclosed technology. FIG. 3 illustrates a per-amino acid classification of the atoms of the amino acids in the reference amino acid sequence of FIG. 2, according to one embodiment of the disclosed technology. 4 is a diagram showing per amino acid assignment of 3D atomic coordinates of alpha carbon atoms classified in FIG. 3 on an amino acid basis, according to one embodiment of the disclosed technology. FIG. FIG. 3 schematically illustrates a process for determining per-voxel distance values, according to one implementation of the disclosed technology. FIG. 3 illustrates an example of a distance channel every 21 amino acids, according to one embodiment of the disclosed technology. 1 is a schematic diagram of a distance channel tensor, according to one implementation of the disclosed technology; FIG. FIG. 3 illustrates one-hot encoding of reference and alternative amino acids from FIG. 2 according to one embodiment of the disclosed technology. 1 is a schematic diagram of voxelized one-hot encoded reference amino acids and voxelized one-hot encoded variants/alternative amino acids according to one embodiment of the disclosed technology; FIG. 8 schematically illustrates a concatenation process for concatenating the distance channel tensor of FIG. 7 and the reference allele tensor voxel by voxel, according to one embodiment of the disclosed technology; FIG. 11 schematically illustrates a concatenation process for concatenating the distance channel tensor of FIG. 7, the reference allele tensor of FIG. 10, and the alternative allele tensor voxel by voxel, according to one implementation of the disclosed technology; FIG. 1 is a flowchart illustrating the process of a system for determining pan-amino acid conservation frequencies of nearest atoms and assigning them to voxels (voxelization), according to one embodiment of the disclosed technology. FIG. 3 shows the closest amino acids from a voxel, according to one embodiment of the disclosed technology. FIG. 2 shows an exemplary multiple sequence alignment of reference amino acid sequences across 99 species, according to one embodiment of the disclosed technology. FIG. 3 is a diagram illustrating an example of determining a pan-amino acid conservation frequency sequence for a specific voxel according to one embodiment of the disclosed technology. FIG. 16 illustrates each pan-amino acid conservation frequency determined for each voxel using the position frequency logic illustrated in FIG. 15, according to one embodiment of the disclosed technology. FIG. 4 illustrates a voxelized voxel-by-voxel evolutionary profile in accordance with one implementation of the disclosed technology. FIG. 3 is a diagram illustrating an example evolutionary profile tensor, according to one implementation of the disclosed technology. 1 is a flowchart illustrating the process of a system for determining the conserved frequency of nearest atoms per amino acid and assigning them to voxels (voxelization), according to one embodiment of the disclosed technology. FIG. 3 illustrates various examples of voxelized annotation channels concatenated with distance channel tensors, according to one implementation of the disclosed technology. FIG. 1 illustrates different combinations and permutations of input channels that can be provided as input to a pathogenicity classifier for determining the pathogenicity of a target variant, according to one embodiment of the disclosed technology. FIGS. 3A and 3B illustrate different methods of calculating the disclosed distance channel according to various implementations of the disclosed technology. FIGS. FIGS. 3A and 3B illustrate different examples of evolutionary channels in accordance with various implementations of the disclosed technology. FIGS. FIGS. 3A and 3B illustrate different examples of annotation channels in accordance with various implementations of the disclosed technology. FIGS. FIGS. 3A and 3B illustrate different examples of structural confidence channels in accordance with various implementations of the disclosed technology; FIGS. FIG. 2 illustrates an example processing architecture for a pathogenicity classifier, in accordance with one implementation of the disclosed technology. FIG. 2 illustrates an example processing architecture for a pathogenicity classifier, in accordance with one implementation of the disclosed technology. FIG. 13 uses PrimateAI as a benchmark model to demonstrate the superiority of the disclosed PrimateAI 3D classification over PrimateAI. FIG. 10 demonstrates the superiority of the disclosed PrimateAI 3D classification over PrimateAI using PrimateAI as a benchmark model; FIG. 10 demonstrates the superiority of the disclosed PrimateAI 3D classification over PrimateAI using PrimateAI as a benchmark model; FIG. 10 demonstrates the superiority of the disclosed PrimateAI 3D classification over PrimateAI using PrimateAI as a benchmark model; 32A and 32B are diagrams illustrating the disclosed efficient voxelization process in accordance with various implementations of the disclosed technology. FIG. 3 is a diagram illustrating how atoms are associated with voxels containing atoms, according to one implementation of the disclosed technology. FIG. 3 illustrates generating a voxel-to-atom mapping from an atom-to-voxel mapping to identify the closest atom to each voxel, according to one implementation of the disclosed technology. How does the disclosed efficient voxelization reduce runtime complexity of 0 (number of atoms) vs. runtime complexity of 0 (number of atoms ^* number of voxels) without using the disclosed efficient voxelization? FIG. How does the disclosed efficient voxelization reduce runtime complexity of 0 (number of atoms) vs. runtime complexity of 0 (number of atoms ^* number of voxels) without using the disclosed efficient voxelization? FIG. 1 illustrates an example computer system that can be used to implement the disclosed techniques. FIG.

The following discussion is presented to enable any person skilled in the art to make and use the disclosed technology, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and embodiments without departing from the spirit and scope of the disclosed technology. It can be applied to various uses. Therefore, the disclosed technology is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

The detailed description of the various embodiments can be better understood when read in conjunction with the accompanying drawings. To the extent that the figures depict diagrams of functional blocks of various implementations, the functional blocks do not necessarily indicate divisions between hardware circuits. Thus, for example, one or more of the functional blocks (e.g., modules, processors, or memories) may be implemented using a single piece of hardware (e.g., a general-purpose signal processor or a block of random access memory, a hard disk, etc.) or multiple pieces of hardware. may be implemented. Similarly, a program may be a standalone program, embedded as a subroutine within an operating system, a function within an installed software package, and so on. It is to be understood that the various embodiments are not limited to the arrangements and instrumentalities shown in the drawings.

The processing engines and databases in the figures designated as modules can be implemented in hardware or software and need not be divided in exactly the same blocks as shown in the figures. Some modules may be implemented on different processors, computers or servers, or even spread among many different processors, computers or servers. In addition, it will be understood that some of the modules can be operated in parallel or in a different order than shown in the figures without affecting the functionality accomplished. The modules in the figures can also be considered as flowchart steps in a method. Also, modules do not necessarily have to have all code located contiguously in memory. Some portions of code can be separated from other portions of code, with code from other modules or other functions located in between.

Determination of Pathogenicity Based on Protein Structure FIG. 1 is a flow diagram illustrating a process 100 of a system for determining pathogenicity of a variant. At step 102, the system's sequence accessor 104 accesses the reference and alternative amino acid sequences. At 112, the system's 3D structure generator 114 generates a 3D protein structure of the reference amino acid sequence. In some embodiments, the 3D protein structure is a homology model of a human protein. In one embodiment, the so-called SwissModel homology modeling pipeline provides a public repository of predicted human protein structures. In another embodiment, so-called HHpred homology modeling uses a tool called Modeler to predict the structure of a target protein from a template structure.

Proteins are represented by collections of atoms and their coordinates in 3D space. Amino acids can have various atoms such as carbon atoms, oxygen (O) atoms, nitrogen (N) atoms, and hydrogen (H) atoms. Atoms can be further classified as side chain atoms and backbone atoms. Backbone carbon atoms can include alpha carbon (C _α ) and beta carbon (C _β ) atoms.

At step 122, the system's coordinate classifier 124 classifies the 3D atomic coordinates of the 3D protein structure on an amino acid basis. In one embodiment, the per-amino acid classification includes assigning 3D atomic coordinates to 21 amino acid categories (including stop or gap amino acid categories). In one example, an amino acid-by-amino acid classification of alpha carbon atoms may list each alpha carbon atom under each of the 21 amino acid categories. In another example, an amino acid-by-amino acid classification of beta carbon atoms can list each beta carbon atom under each of the 21 amino acid categories.

In yet another example, an amino acid-by-amino acid classification of oxygen atoms can list each oxygen atom under each of the 21 amino acid categories. In yet another example, the amino acid-by-amino acid classification of nitrogen atoms can list each nitrogen atom under each of the 21 amino acid categories. In yet another example, an amino acid-by-amino acid classification of hydrogen atoms may list hydrogen atoms under each of the 21 amino acid categories.

Those skilled in the art will appreciate that in various implementations, classification by amino acid can include subsets of the 21 amino acid categories and subsets of different atomic elements.

At step 132, the system's voxel grid generator 134 instantiates a voxel grid. The voxel grid can have any resolution, for example 3x3x3, 5x5x5, 7x7x7, etc. The voxels in the voxel grid can be of any size, such as 1 angstrom (Å) on each side, 2 Å on each side, 3 Å on each side, etc. Those skilled in the art will understand that these exemplary dimensions refer to cubic dimensions, since voxels are cubic. Those skilled in the art will also appreciate that these example dimensions are non-limiting and that voxels can have any cubic dimensions.

In step 142, the system's voxel grid centerer 144 centers the voxel grid on a reference amino acid that experiences the target variant at the amino acid level. In one embodiment, the voxel grid is centered at the atomic coordinates of a particular atom of the reference amino acid that experiences the target variant, such as the 3D atomic coordinates of the alpha carbon atom of the reference amino acid that experiences the target variant.

Distance Channels Voxels within a voxel grid can have multiple channels (or features). In one implementation, the voxels in the voxel grid have multiple distance channels (eg, 21 distance channels for each of the 21 amino acid categories, including stop or gap amino acid categories). At step 152, the system's distance channel generator 154 generates per-amino acid distance channels for voxels in the voxel grid. Distance channels are generated independently for each of the 21 amino acid categories.

For example, consider the alanine (A) amino acid category. Further, consider, for example, that the voxel grid is 3x3x3 in size and has 27 voxels. Then, in one implementation, the alanine distance channels each include 27 distance values for 27 voxels in the voxel grid. The 27 distance values in the alanine distance channel are measured from the center of each of the 27 voxels in the voxel grid to the respective nearest atom in the alanine amino acid category.

In one example, the alanine amino acid category includes only alpha carbon atoms, so the nearest atom is the alanine alpha carbon atom that is closest to each of the 27 voxels in the voxel grid. In another example, the alanine amino acid category includes only beta carbon atoms, so the nearest atom is the alanine beta carbon atom that is closest to each of the 27 voxels in the voxel grid.

In yet another example, the alanine amino acid category includes only oxygen atoms, so the nearest atom is the alanine oxygen atom that is closest to each of the 27 voxels in the voxel grid. In yet another example, the alanine amino acid category includes only nitrogen atoms, so the nearest atom is the alanine nitrogen atom that is closest to each of the 27 voxels in the voxel grid. In yet another example, the alanine amino acid category contains only hydrogen atoms, so the nearest atom is the alanine hydrogen atom that is closest to each of the 27 voxels in the voxel grid.

Similar to the alanine distance channel, distance channel generator 154 generates a distance channel (i.e., a set of voxel-wise distance values) for each of the remaining amino acid categories. In other embodiments, distance channel generator 154 generates distance channels only for a subset of 21 amino acid categories.

In other embodiments, the selection of closest atoms is not limited to a particular atom type. That is, within the amino acid category of interest, the closest atom to a given voxel is selected regardless of the atomic element of the nearest atom, and the distance value of the given voxel is calculated for inclusion in the distance channel of the amino acid category of interest. Ru.

In yet other embodiments, distance channels are generated on an atomic element basis. Instead of or in addition to having distance channels for amino acid categories, distance values can be generated for atomic element categories regardless of the amino acid to which the atom belongs. For example, consider that the atoms of the amino acids in the reference amino acid sequence span seven atomic elements: carbon, oxygen, nitrogen, hydrogen, calcium, iodine, and sulfur. Voxels in the voxel grid are then configured to have seven distance channels, such that each of the seven distance channels specifies the distance to the nearest atom only within the corresponding atomic element category. has a per-voxel distance value of . In other implementations, distance channels may be generated for only a subset of seven atomic elements. In yet other implementations, the atomic element categories and distance channel generation can be further stratified into variations of the same atomic element, eg, alpha carbon (C _α ) atoms and beta carbon (C _β ) atoms.

In yet other embodiments, distance channels can be generated on an atom type basis, such as distance channels for side chain atoms only and distance channels for backbone atoms only.

The closest atom can be searched within a predetermined maximum scan radius (eg, 6 angstroms (Å)) from the voxel center. Also, multiple atoms may be closest to the same voxel in the voxel grid.

The distance is calculated between the 3D coordinates of the voxel center and the 3D atomic coordinates of the atom. Distance channels are also generated using a voxel grid that is centered at the same location (e.g., centered at the 3D atomic coordinates of the alpha carbon atom of the reference amino acid experiencing the target variant).

The distance can be a Euclidean distance. The distance can also be parameterized by atomic size (or atomic influence) (eg, by using the Lennard-Jones potential and/or van der Waals atomic radius of the atom in question). The distance value can also be normalized by the maximum scan radius or by the maximum observed distance value of the furthest nearest atom in the amino acid category of interest or the atomic element category of interest or the atom type category of interest. In some implementations, distances between voxels and atoms are calculated based on the polar coordinates of the voxels and atoms. Polar coordinates are parameterized by the angle between the voxel and the atom. In one implementation, this angular information is used to generate the angular channel of the voxel (ie, independent of the distance channel). In some implementations, the angle between the closest atom and a neighboring atom (eg, a backbone atom) can be used as a feature encoded using voxels.

Reference Allele and Alternative Allele Channels Voxels within a voxel grid can also have reference allele and alternative allele channels. In step 162, the system's one-hot encoder 164 generates a reference one-hot encoding of the reference amino acid in the reference amino acid sequence and an alternate one-hot encoding of the alternate amino acid in the alternate amino acid sequence. The reference amino acid undergoes targeted mutations. Alternative amino acids are targeted variants. The reference amino acid and the alternative amino acid are located at the same position in the reference amino acid sequence and the alternative amino acid sequence, respectively. The reference amino acid sequence and the alternative amino acid sequence have the same position-by-position amino acid composition, with one exception. An exception is a position with a reference amino acid in a reference amino acid sequence and an alternative amino acid in an alternative amino acid sequence.

At step 172, a concatenator 174 of the system concatenates the per-amino acid distance channel with the reference and alternative one-hot encodings. In another embodiment, concatenator 174 concatenates the per-atomic element distance channel and the reference one-hot encoding and the alternative one-hot encoding. In yet another embodiment, concatenator 174 concatenates the distance channel for each atom type with the reference one-hot encoding and the alternative one-hot encoding.

In step 182, the system runtime logic 184 passes the concatenated per-amino acid/per-atomic element/per-atom type distance channels and reference and alternative one-hot encodings through a pathogenicity classifier (virulence decision engine). are processed to determine the pathogenicity of the target variant, which is then inferred as determining the pathogenicity of the underlying nucleotide variant generating the target variant at the amino acid level. A pathogenicity classifier is trained using labeled datasets of benign and pathogenic variants, for example, using an error backpropagation algorithm. Further details regarding labeled datasets of benign and pathogenic variants, as well as exemplary architectures and training of pathogenicity classifiers, can be found in commonly-owned U.S. patent application Ser. No. 16/160,903; No. 986, No. 16/160,968, and No. 16/407,149.

FIG. 2 schematically depicts a reference amino acid sequence 202 of protein 200 and an alternative amino acid sequence 212 of protein 200. Protein 200 contains N amino acids. Amino acid positions in protein 200 are labeled 1, 2, 3,...,N. In the illustrated example, position 16 is the position that experiences an amino acid variant 214 (mutation) caused by the underlying nucleotide variant. For example, for reference amino acid sequence 202, position 1 has the reference amino acid phenylalanine (F), position 16 has the reference amino acid glycine (G) 204, and position N (e.g., the last amino acid of sequence 202) has the reference amino acid phenylalanine (F), and position N (e.g., the last amino acid of sequence 202) It has the amino acid leucine (L). Although not illustrated for clarity, the remaining positions in reference amino acid sequence 202 contain various amino acids in order specific to protein 200. Alternative amino acid sequence 212 is the same as reference amino acid sequence 202 except for variant 214 at position 16, which contains the alternative amino acid alanine (A) 214 in place of reference amino acid glycine (G) 204.

FIG. 3 shows a per-amino acid classification of the atoms of the amino acids in the reference amino acid sequence 202, also referred to herein as "atomic classification 300." Certain types of amino acids among the 20 naturally occurring amino acids listed in column 302 may be repeated in proteins. That is, a particular type of amino acid may occur more than once in a protein. A protein may also have some undetermined amino acids classified by the 21st stop or gap amino acid category. The right column of Figure 3 contains counts of alpha carbon (C _α ) atoms from different amino acids.

Specifically, FIG. 3 shows a classification of the alpha carbon (C _α ) atoms of the amino acids in the reference amino acid sequence 202 by amino acid. Column 308 of FIG. 3 lists the total number of alpha carbon atoms observed for the reference amino acid sequence 202 in each of the 21 amino acid categories. For example, column 308 lists the 11 alpha-carbon atoms observed for the alanine (A) amino acid category. Since each amino acid has only one alpha carbon atom, this means that alanine occurs 11 times in the reference amino acid sequence 202. In another example, arginine (R) occurs 35 times in reference amino acid sequence 202. The total number of alpha carbon atoms across the 21 amino acid categories is 828.

FIG. 4 shows the assignment of the 3D atomic coordinates of the alpha carbon atoms of the reference amino acid sequence 202 for each amino acid based on the atomic classification 300 of FIG. 3. This is referred to herein as "atomic coordinate bucketing 400." In FIG. 4, lists 404-440 tabulate the 3D atomic coordinates of alpha carbon atoms bucketed into each of the 21 amino acid categories.

In the illustrated embodiment, bucketing 400 of FIG. 4 follows classification 300 of FIG. 3. For example, in FIG. 3, the alanine amino acid category has 11 alpha carbon atoms; therefore, in FIG. 4, the alanine amino acid category has 11 3D atoms of the corresponding 11 alpha carbon atoms from FIG. Has coordinates. This classification-to-bucketing logic also flows from FIG. 3 to FIG. 4 for other amino acid categories. However, this classification-to-bucketing logic is only for representational purposes; in other implementations, the disclosed technique uses classification 300 and bucketing to locate the nearest atom per voxel. 400; fewer, additional, or different steps may be performed. For example, in some embodiments, the disclosed techniques include sorting criteria (e.g., per amino acid, per atomic element, per atom type), predetermined maximum scan radius, and distance type (e.g., Euclidean, By using a sorting and searching algorithm that returns the nearest atom per voxel from one or more databases in response to a search query configured to accept query parameters such as Mahalanobis, normalization, denormalization), etc. , the nearest atom for each voxel can be located. In various embodiments of the disclosed technology, multiple sorting and search algorithms from the current or future technical field may similarly be used by those skilled in the art to locate the nearest atom per voxel. I can do it.

In Figure 4, the 3D atomic coordinates are represented by Cartesian coordinates x, y, z, but any type of coordinate system may be used, such as spherical or cylindrical coordinates, and the claimed subject matter is not limited to. In some embodiments, one or more databases may include information regarding the 3D atomic coordinates of alpha carbon atoms and other atoms of amino acids in proteins. Such databases may be searchable by specific proteins.

As mentioned above, voxels and voxel grids are 3D entities. However, for clarity, the drawings show voxels and voxel grids in two-dimensional (2D) format, and the description discusses the same. For example, a 3x3x3 voxel grid of 27 voxels is shown and described herein as a 3x3 2D pixel grid with 9 2D pixels. Those skilled in the art will appreciate that the 2D format is used for representational purposes only and is intended to cover its 3D counterpart (i.e., 2D pixels represent 3D voxels and 2D pixel grids represent 3D voxel grids). you will understand. Also, the drawings are not to scale. For example, a voxel of size 2 Angstroms (Å) is described using a single pixel.

Voxel-wise Distance Calculation FIG. 5 schematically depicts the process of determining a voxel-wise distance value, also referred to herein as "voxel-wise distance calculation 500." In the illustrated example, per-voxel distance values are calculated only for the alanine (A) distance channel. However, as discussed above for each Figure 1, we performed the same distance calculation logic for each of the 21 amino acid categories to generate distance channels for each of the 21 amino acids, and other Further extensions can be made to atomic types and other atomic elements such as oxygen, nitrogen, and hydrogen. In some implementations, atoms are randomly rotated before distance calculation to make the training of the pathogenicity classifier invariant to atom orientation.

In FIG. 5, the voxel grid 522 is indexed (1,1), (1,2), (1,3), (2,1), (2,2), (2,3), (3,1 ), (3,2), and (3,3). Voxel grid 522 is centered, for example, at the 3D atomic coordinates 532 of the alpha carbon atom of the glycine (G) amino acid at position 16 of reference amino acid sequence 202, which corresponds to alternative amino acid sequence 212, as described above with respect to FIG. , because position 16 experiences a mutation that mutates the glycine (G) amino acid to the alanine (A) amino acid. Further, the center of voxel grid 522 coincides with the center of voxel (2, 2).

A centered voxel grid 522 is used for voxel-wise distance calculations for each of the 21 amino acid-wise distance channels. For example, starting with the alanine (A) distance channel, measure the distance between the 3D coordinates of the center of each of the 9 voxels 514 and the 3D atomic coordinates 402 of the 11 alanine alpha carbon atoms, and Locate the nearest alanine alpha carbon atom for each voxel 514 of . The nine distance values for the nine distances between the nine voxels 514 and their respective nearest alanine alpha carbon atoms are then used to construct the alanine distance channel. The resulting alanine distance channel places the nine alanine distance values in the same order as the nine voxels 514 in the voxel grid 522.

The above process is performed for each of the 21 amino acid categories. For example, the center voxel grid 522 is similarly used to calculate the arginine (R) distance channel, thus providing the 3D coordinates of the center of each of the 9 voxels 514 and the 3D atoms of the 35 arginine alpha carbon atoms. The distance between coordinates 404 is measured to locate the nearest arginine alpha carbon atom for each of the nine voxels 514. The nine distance values for the nine distances between the nine voxels 514 and their respective nearest arginine alpha carbon atoms are then used to construct the arginine distance channel. The resulting arginine distance channel places the nine arginine distance values in the same order as the nine voxels 514 in the voxel grid 522. A distance channel of every 21 amino acids is encoded voxel by voxel to form a distance channel tensor.

Specifically, in the illustrated example, distance 512 is between the center of voxel (1,1) of voxel grid 522 and the nearest alpha carbon (C _α ) atom, which is a C α ^A5 atom in list 402. It is. Therefore, the value assigned to voxel (1,1) is distance 512. In another example, the Cα ^A4 atom is the _Cα atom closest to the center of voxel (1,2). Therefore, the value assigned to voxel (1,2) is the distance between the center of voxel (1,2) and the Cα ^A4 atom. In yet another example, the Cα ^A6 atom is the _Cα atom closest to the center of voxel (2,1). Therefore, the value assigned to voxel (2,1) is the distance between the center of voxel (2,1) and the Cα ^A6 atom. In yet another example, the Cα ^A6 atom is also the _Cα atom closest to the center of voxels (3,2) and (3,3). Therefore, the value assigned to voxel (3,2) is the distance between the center of voxel (3,2) and the Cα ^A6 atom, and the value assigned to voxel (3,3) is the distance between the center of voxel (3,2) and the Cα A6 atom; 3,3) and the Cα ^A6 atom. In some implementations, the distance value assigned to voxel 514 may be a normalized distance. For example, the distance value assigned to voxel (1,1) may be distance 512 divided by maximum distance 502 (predetermined maximum scan radius). In some implementations, the closest atomic distance may be a Euclidean distance, where the closest atomic distance is normalized by dividing the Euclidean distance by the largest nearest atomic distance (e.g., maximum distance 502, etc.). may be converted into

As noted above, for amino acids having an alpha carbon atom, the distance can be the distance of the nearest alpha carbon atom from the corresponding voxel center to the nearest alpha carbon atom of the corresponding amino acid. Additionally, for amino acids having a beta carbon atom, the distance can be the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding amino acid. Similarly, for amino acids having backbone atoms, the distance can be the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding amino acid. Similarly, for amino acids having side chain atoms, the distance can be the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding amino acid. In some implementations, the distance can additionally/alternatively include the distance to the second, third, fourth nearest atom, etc.

Per Amino Acid Distance Channels FIG. 6 shows an example of 21 per amino acid distance channels 600. Each column of FIG. 6 corresponds to a respective one of the 21 per amino acid distance channels 602-642. Each per amino acid distance channel includes a distance value for each of the voxels 514 of the voxel grid 522. For example, the per amino acid distance channel 602 for alanine (A) includes a distance value for each of the voxels 514 of the voxel grid 522. As discussed above, the voxel grid 522 is a 3D grid of volume 3×3×3 and includes 27 voxels. Similarly, although FIG. 6 shows the voxels 514 in two dimensions (e.g., 9 voxels in a 3×3 grid), each per amino acid distance channel may include a distance value for each of the 27 voxels for the 3×3×3 voxel grid.

Directionality Encoding In some embodiments, the disclosed technology uses directionality parameters to specify the directionality of reference amino acids within reference amino acid sequence 202. In some embodiments, the disclosed technology uses directionality parameters to specify the directionality of alternative amino acids within alternative amino acid sequence 212. In some embodiments, the disclosed technology uses directional parameters to specify locations within protein 200 that experience targeted variants at the amino acid level.

As mentioned above, all distance values in distance channels 602-642 for every 21 amino acids are measured from the respective nearest atom to voxel 514 in voxel grid 522. These closest atoms are derived from one of the reference amino acids in reference amino acid sequence 202. These original reference amino acids that contain the closest atoms can be classified into two categories: (1) the original reference amino acids that precede the reference amino acid 204 that experience variants in the reference amino acid sequence 202 and ( 2) The original reference amino acid following the reference amino acid 204 that experiences a variant in the reference amino acid sequence 202. The original reference amino acids in the first category can be referred to as preceding reference amino acids. The original reference amino acid in the second category can be referred to as the subsequent reference amino acid.

Directional parameters are applied to distance values in distance channels 602-642 for every 21 amino acids measured from the nearest atom derived from the preceding reference amino acid. In one implementation, the directional parameter is multiplied by such distance value. The directionality parameter can be any number, such as -1.

As a result of the application of the directional parameter, the distance channels 600 for every 21 amino acids contain several distances that indicate to the pathogenicity classifier which end of the protein 200 is the starting end and which end is the ending end. Contains value. This also allows the pathogenicity classifier to reconstruct protein sequences from the 3D protein structure information provided by the distance channel as well as the reference and allele channels.

Distance Channel Tensor FIG. 7 is a schematic diagram of a distance channel tensor 700. Distance channel tensor 700 is a voxelized representation of per-amino acid distance channel 600 from FIG. In the distance channel tensor 700, the distance channels 602-642 for every 21 amino acids are concatenated voxel by voxel like the RGB channels of a color image. The voxelized dimensionality of the distance channel tensor 700 is 21x3x3x3 (where 21 indicates the 21 amino acid categories and 3x3x3 represents the 3D voxel with 27 voxels). grid). However, FIG. 7 is a 2D depiction with dimensions 21×3×3.

One-Hot Encoding FIG. 8 shows one-hot encoding 800 of reference amino acid 204 and alternative amino acid 214. In FIG. 8, the left column is a one-hot encoding 802 of the reference amino acid glycine (G) 204, where 1 is for the glycine amino acid category and 0 is for all other amino acid categories. In FIG. 8, the right column is a one-hot encoding 804 of the variant/alternative amino acid alanine (A) 214, where 1 is for the alanine amino acid category and 0 is for all other amino acid categories.

FIG. 9 is a schematic diagram of a voxelized one-hot encoded reference amino acid 902 and a voxelized one-hot encoded variant/alternative amino acid 912. Voxelized one-hot encoded reference amino acid 902 is a voxelized representation of one-hot encoded 802 of reference amino acid glycine (G) 204 from FIG. Voxelized one-hot encoded alternative amino acid 912 is a voxelized representation of one-hot encoded 804 of variant/alternative amino acid alanine (A) 214 from FIG. The voxelization dimension number of the voxelized one-hot encoding reference amino acid 902 is 21×1×1×1 (here, 21 indicates 21 amino acid categories). However, FIG. 9 is a 2D depiction with dimensions 21×1×1. Similarly, the voxelization dimensionality of the voxelized one-hot encoded alternative amino acids 912 is 21×1×1×1 (where 21 indicates the 21 amino acid categories). However, FIG. 9 is a 2D depiction with dimensions 21×1×1.

Reference Allele Tensor FIG. 10 schematically depicts a concatenation process 1000 that concatenates the distance channel tensor 700 of FIG. 7 and the reference allele tensor 1004 voxel by voxel. Reference allele tensor 1004 is a voxel-wise collection (repeat/cloned/replicated) of voxelized one-hot encoded reference amino acids 902 from FIG. That is, multiple copies of the voxelized one-hot encoded reference amino acid 902 cause the reference allele tensor 1004 to correspond to a corresponding copy of the voxelized one-hot encoded reference amino acid 910 for each voxel 514 in the voxel grid 522. are connected voxel by voxel to each other with respect to the spatial arrangement of voxels 514 within voxel grid 522 such that .

Concatenation process 1000 generates concatenation tensor 1010. The voxelized dimensionality of the reference allele tensor 1004 is 21x3x3x3 (where 21 indicates the 21 amino acid categories and 3x3x3 is the 3D dimension with 27 voxels). (showing voxel grid). However, FIG. 10 is a 2D depiction of a reference allele tensor 1004 with dimensionality 21x3x3. The number of voxelized dimensions of the connected tensor 1010 is 42×3×3×3. However, FIG. 10 is a 2D depiction of a connected tensor 1010 with dimensionality 42x3x3.

Alternate Allele Tensor FIG. 11 schematically depicts a concatenation process 1100 that concatenates the distance channel tensor 700 of FIG. 7, the reference allele tensor 1004 of FIG. 10, and the alternate allele tensor 1104 voxel by voxel. Alternative allele tensor 1104 is a voxel-wise collection (repeat/cloned/replicated) of voxelized one-hot encoded alternative amino acids 912 from FIG. That is, multiple copies of the voxelized one-hot encoded alternative amino acids 912 cause the alternative allele tensor 1104 to represent corresponding copies of the voxelized one-hot encoded alternative amino acids 910 for each voxel 514 in the voxel grid 522. are connected voxel by voxel to each other according to the spatial arrangement of the voxels 514 within the voxel grid 522 such that

Concatenation process 1100 generates concatenation tensor 1110. The voxelized dimensionality of the alternative allele tensor 1104 is 21x3x3x3 (where 21 indicates the 21 amino acid categories and 3x3x3 is the 3D dimension with 27 voxels). (showing voxel grid). However, FIG. 11 is a 2D depiction of an alternative allele tensor 1104 with dimensionality 21x3x3. The number of voxelized dimensions of the connected tensor 1110 is 63×3×3×3. However, FIG. 11 is a 2D depiction of a connected tensor 1110 with dimensionality 63x3x3.

In some implementations, runtime logic 184 processes concatenated tensor 1110 through a pathogenicity classifier to determine the pathogenicity of variant/alternative amino acid alanine (A) 214, which in turn It is speculated that the pathogenicity of the underlying nucleotide variant producing the amino acid/alternative amino acid alanine (A)214.

Evolutionary Conservation Channel Predicting the functional outcome of a variant depends, at least in part, on whether amino acids important to a protein family are conserved through evolution by negative selection (i.e., amino acid changes at these sites are mutations at these sites have a high likelihood of being pathogenic (causing disease) in humans. Generally, homologous sequences of a target protein are collected and aligned, and a conservation metric is calculated based on the weighted frequencies of different amino acids observed at the target position in the alignment.

Accordingly, the disclosed technique concatenates distance channel tensor 700, reference allele tensor 1004, and alternative allele tensor 1104 with an evolutionary channel. An example of an evolutionary channel is pan-amino acid conservation frequency. Another example of an evolutionary channel is the conserved frequency of each amino acid.

In some implementations, evolutionary channels are constructed using position-specific weight matrices (PWM). In other implementations, the evolutionary channel is constructed using a position specific frequency matrix (PSFM). In yet other embodiments, evolutionary channels are constructed using computational tools such as SIFT, PolyPhen, and PANTHER-PSEC. In yet other embodiments, the evolutionary channel is a conserved channel based on evolutionary conservation. Conservation is relevant because it also reflects the effects of negative selection that acted to prevent evolutionary change at a given site in a protein.

Pan-Amino Acid Evolutionary Profiles FIG. 12 is a flowchart illustrating a process 1200 of a system for determining and assigning pan-amino acid conservation frequencies of nearest atoms to voxels (voxelization), according to one embodiment of the disclosed technology. . 12, 13, 14, 15, 16, 17, and 18 will be described in parallel.

At step 1202 , the system's similar sequence finder 1204 searches for amino acid sequences that are similar (homologous) to the reference amino acid sequence 202 . Similar amino acid sequences can be selected from multiple species such as primates, mammals, and vertebrates.

At step 1212, the system's aligner 1214 aligns the reference amino acid sequence 202 position by position with similar amino acid sequences, ie, the aligner 1214 performs a multiple sequence alignment. FIG. 14 shows an exemplary multiple sequence alignment 1400 of reference amino acid sequences 202 across 99 species. In some implementations, the multiple sequence alignment 1400 includes, for example, a first position frequency matrix 1402 for primates, a second position frequency matrix 1412 for mammals, and a third position frequency matrix for primates. It can be partitioned to generate a location frequency matrix 1422. In other implementations, a single location frequency matrix is generated across 99 species.

At step 1222, the system's pan-amino acid conservation frequency calculator 1224 determines the pan-amino acid conservation frequency of the reference amino acid in the reference amino acid sequence 202 using multiple sequence alignments.

At step 1232, the system's closest atom finder 1234 finds the closest atom to voxel 514 in voxel grid 522. In some embodiments, the voxel-by-voxel nearest atom search may not be limited to any particular amino acid category or atom type. That is, the closest atoms per voxel can be selected across amino acid categories and amino acid types, as long as they are the atoms closest to the respective voxel center. In other embodiments, the voxel-by-voxel nearest atom search may include only certain atomic elements such as oxygen, nitrogen, and hydrogen, or only alpha carbon atoms, or only beta carbon atoms, or only side chain atoms, or only side chain atoms, or backbone atoms. It can be limited to only specific atomic categories, such as atoms only.

In step 1242, the system's amino acid selector 1244 selects the reference amino acid in the reference amino acid sequence 202 that contains the closest atom identified in step 1232. Such a reference amino acid can be referred to as the closest reference amino acid. FIG. 13 shows an example of locating the nearest atom 1302 to voxel 514 in voxel grid 522 and mapping the nearest reference amino acid 1312 containing the nearest atom 1302 to voxel 514 in voxel grid 522, respectively. . This is identified in FIG. 13 as "Voxel to nearest amino acid mapping 1300".

In step 1252, the system's voxelizer 1254 voxels the pan-amino acid conservation frequency of the nearest reference amino acid. FIG. 15 shows an example of determining the pan-amino acid conservation frequency sequence for the first voxel (1,1) in the voxel grid 522, also referred to herein as "voxel-wise evolutionary profile determination 1500."

Referring to FIG. 13, the closest reference amino acid mapped to the first voxel (1,1) is the aspartic acid (D) amino acid at position 15 in reference amino acid sequence 202. A multiple sequence alignment between the reference amino acid sequence 202 and, for example, 99 homologous amino acid sequences at position 15 is analyzed. Such position-specific and cross-species analyzes indicate how many examples of amino acids from each of the 21 amino acid categories are found across 100 aligned amino acid sequences (i.e., reference amino acid sequence 202 + 99 homologous amino acid sequences). We will reveal whether you will be discovered in 15th place.

In the example shown in Figure 15, the aspartic acid (D) amino acid is found at position 15 in 96 out of 100 aligned amino acid sequences. Therefore, aspartate amino acid category 1504 is assigned a pan-amino acid conservation frequency of 0.96. Similarly, in the example shown, the valine (V) amino acid is found at position 15 in four out of 100 aligned amino acid sequences. Therefore, the valic acid amino acid category 1514 is assigned a pan-amino acid conservation frequency of 0.04. Since no examples of amino acids from other amino acid categories are detected at position 15, the remaining amino acid categories are assigned a pan-amino acid conservation frequency of 0. In this way, each of the 21 amino acid categories is assigned a respective pan-amino acid conservation frequency, which can be encoded in the pan-amino acid conservation frequency array 1502 for the first voxel (1,1). can.

FIG. 16 illustrates each of the voxels 514 determined for each of the voxels 514 in the voxel grid 522 using the location frequency logic described in FIG. Pan-amino acid conservation frequencies 1612 to 1692 are shown.

The voxel-wise evolutionary profile 1602 is then used by the voxelizer 1254 to generate the voxelized voxel-wise evolutionary profile 1700 shown in FIG. In many cases, each of the voxels 514 within the voxel grid 522 has a different pan-amino acid conservation frequency sequence, and thus the voxels map regularly to different nearest atoms and therefore to different nearest reference amino acids; Each voxel has a different voxelization evolutionary profile. Of course, if two or more voxels have the same nearest atom and therefore the same nearest reference amino acid, then the same pan-amino acid conserved frequency sequence and the same voxel-by-voxel voxelization evolutionary profile will result in two or more voxels are assigned to each of the following.

Figure 18 shows an example of an evolutionary profile tensor 1800 in which the voxelized per-voxel evolutionary profiles 1700 are connected to each other voxel-wise according to the spatial arrangement of the voxels 514 in the voxel grid 522. The voxelized dimensionality of the evolutionary profile tensor 1800 is 21x3x3x3 (where 21 indicates 21 amino acid categories and 3x3x3 indicates a 3D voxel grid with 27 voxels). However, Figure 18 is a 2D depiction of the evolutionary profile tensor 1800 with dimensionality 21x3x3.

At step 1262, concatenator 174 concatenates evolutionary profile tensor 1800 with distance channel tensor 700 voxel by voxel. In some implementations, evolutionary profile tensor 1800 is concatenated voxel-wise with connectivity tensor 1110 to generate an additional connectivity tensor (not shown) of dimensionality 84x3x3x3.

In step 1272, the runtime logic 184 processes the further concatenated tensor of dimensions 84x3x3x3 through a pathogenicity classifier to determine the pathogenicity of the target variant, which is then inferred as the pathogenicity determination of the underlying nucleotide variant that generates the target variant at the amino acid level.

Evolutionary Profiles Per Amino Acid Figure 19 is a flow diagram showing a process 1900 of a system for determining the conservation frequency per amino acid of nearest atoms and assigning them to voxels (voxelizing). In Figure 19, steps 1202 and 1212 are the same as in Figure 12.

At step 1922, the system's per-amino acid conservation frequency calculator 1924 determines the per-amino acid conservation frequency of the reference amino acid in the reference amino acid sequence 202 using multiple sequence alignments.

At step 1932, the system's closest atom finder 1934 finds, for each voxel 514 in voxel grid 522, the 21 closest atoms across each of the 21 amino acid categories. Each of the 21 nearest atoms is different from each other because they are selected from different amino acid categories. This leads to the selection of the 21 unique closest reference amino acids for a particular voxel, which in turn leads to the generation of 21 unique position frequency matrices for a particular voxel, which in turn This leads to the determination of the conservation frequency of each of the 21 unique amino acids for a particular voxel.

In step 1942, the system's amino acid selector 1944 selects, for each voxel 514 in the voxel grid 522, the 21 reference amino acids in the reference amino acid sequence 202 that contain the 21 nearest atoms identified in step 1932. select. Such a reference amino acid can be referred to as the closest reference amino acid.

In step 1952, the system's voxelizer 1954 voxelizes the pen-amino acid conservation frequencies of the 21 closest reference amino acids identified for the particular voxel in step 1942. The 21 closest reference amino acids are necessarily located at 21 different positions in the reference amino acid sequence 202 because they correspond to different underlying nearest atoms. Therefore, for a particular voxel, 21 position frequency matrices can be generated for the 21 nearest reference amino acids. The 21 position frequency matrices can be generated across multiple species whose homologous amino acid sequences are aligned position by position with the reference amino acid sequence 202, as described above with respect to FIGS. 12-15.

Using the 21 position frequency matrices, 21 position-specific conservation scores can then be calculated for the 21 closest reference amino acids identified for a particular voxel. These 21 position-specific conservation scores form the pen-amino acid conservation frequency for a specific voxel, similar to the pan-amino acid conservation frequency array 1502 in FIG. However, because the 21 closest reference amino acids across the 21 amino acid categories necessarily have different positions resulting in different position frequency matrices and thereby different conservation frequencies for each amino acid, sequence 1502 has many 0 entries. , where each element (feature) in the conserved frequency array for each amino acid has a value (eg, a floating point number).

The above process is performed for each of the voxels 514 in the voxel grid 522, and the resulting per-amino acid conservation frequency per voxel is similar to the pan-amino acid conservation frequency described with respect to FIGS. 12-18. Voxelized, tensorized, concatenated and processed for pathogenicity determination.

Annotation Channels FIG. 20 shows various examples of voxelized annotation channels 2000 coupled with distance channel tensor 700. In some embodiments, the voxelized annotation channels are configured to have different protein annotations, e.g., whether the amino acid (residue) is part of a transmembrane region, signal peptide, active site, or any other binding site. (Pei P, Zhang A: A Topological Measurement for Weighted Protein Interaction Network. CSB 2005, 26 8-278). Additional examples of annotation channels can be found in the Specific Implementations section below and in the claims.

Voxelized annotation channels allow voxels to have the same annotation sequences, such as voxelized reference allele and alternative allele sequences (e.g., annotation channels 2002, 2004, 2006), or may have respective annotation arrays such as a voxelized per-voxel evolutionary profile 1700 (e.g. annotation channels 2012, 2014, 2016 (as indicated by different colors)) for each voxel. It will be placed in

The annotation channels are voxelized, tensorized, concatenated, and processed for pathogenicity determination similar to the pan-amino acid conservation frequencies described with respect to FIGS. 12-18.

Structural Confidence Channels The disclosed techniques can also connect various voxelized structural confidence channels with the distance channel tensor 700. Some examples of structural confidence channels include GMQE scores (provided by SwissModel); B factor; temperature factor sequence of the homology model (indicating how well the residues satisfy the (physical) constraints in the protein structure); the normalized number of aligned template proteins relative to the residue closest to the center of the voxel ( Alignments provided by HHpred, e.g., voxels are closest to the residues that 3 of the 6 template structures align, meaning the feature has the value 3/6 = 0.5); minimum, maximum , and the average TM score; and the predicted TM score of the template protein structure that aligns with the residue closest to the voxel (continuing the example above, the three template structures have TM scores of 0.5, 0.5, and 1.5 , the minimum is 0.5, the average is 2/3, and the maximum is 1.5). A TM score can be provided for each protein template by HHpred. Additional examples of structure confidence channels can be found in the Specific Implementations section below and in the claims.

The voxelized structural confidence channel allows voxels to have the same structural confidence sequences, such as the voxelized reference allele and alternative allele sequences, or Each voxel is arranged such that each voxel can have its own structural confidence array, such as an evolutionary profile 1700 for each voxel.

Structural confidence channels are voxelized, tensorized, concatenated, and processed for pathogenicity determination similar to the pan-amino acid conservation frequencies described with respect to FIGS. 12-18.

Pathogenicity Classifier FIG. 21 illustrates different combinations and permutations of input channels that can be provided as inputs 2102 to a pathogenicity classifier 2108 for pathogenicity determination 2106 of a target variant. One of the inputs 2102 can be a distance channel 2104 generated by a distance channel generator 2272. FIG. 22 shows different ways to calculate the distance channel 2104. In one implementation, distance channels 2104 are generated based on distances 2202 between voxel centers and atoms across multiple atomic elements, regardless of amino acid. In some implementations, distance 2202 is normalized by the maximum scan radius to produce normalized distance 2202a. In another embodiment, the distance channel 2104 is generated based on the distance 2212 between the voxel center and the alpha carbon atom on an amino acid basis. In some implementations, distance 2212 is normalized by the maximum scan radius to generate normalized distance 2212a. In another embodiment, the distance channel 2104 is generated based on the distance 2222 between the voxel center and the beta carbon atom on an amino acid basis. In some implementations, distance 2222 is normalized by the maximum scan radius to produce normalized distance 2222a. In another embodiment, the distance channel 2104 is generated based on the distance 2232 between the voxel center and the side chain atom on an amino acid basis. In some implementations, distance 2232 is normalized by the maximum scan radius to produce normalized distance 2232a. In another embodiment, the distance channel 2104 is generated based on the distance 2242 between the voxel center and the backbone atoms on an amino acid basis. In some implementations, distance 2242 is normalized by the maximum scan radius to produce normalized distance 2242a. In yet another embodiment, the distance channel 2104 is generated based on the distance 2252 (one feature) between the voxel center and the respective nearest atom, regardless of atom type and amino acid type. In yet another embodiment, the distance channel 2104 is generated based on the distance 2262 (one feature) between the voxel center and the atom from the non-canonical amino acid. In some implementations, distances between voxels and atoms are calculated based on the polar coordinates of the voxels and atoms. Polar coordinates are parameterized by the angle between the voxel and the atom. In one implementation, this angular information is used to generate the angular channel of the voxel (ie, independent of the distance channel). In some implementations, the angle between the closest atom and a neighboring atom (eg, a backbone atom) can be used as a feature encoded using voxels.

Another of the inputs 2102 can be features 2114 that indicate missing atoms within a specified radius.

Another of the inputs 2102 can be a one-hot encoding 2124 of reference amino acids. Another of the inputs 2102 can be one-hot encoding 2134 of variant/alternative amino acids.

Another of the inputs 2102 can be an evolutionary channel 2144 generated by the evolutionary profile generator 2372 shown in FIG. 23. In one embodiment, evolutionary channels 2144 can be generated based on pan-amino acid conservation frequencies 2302. In one embodiment, evolutionary channels 2144 can be generated based on pan-amino acid conservation frequencies 2312.

Another of the inputs 2102 can be features 2154 that indicate missing residues or missing evolutionary profiles.

Another of the inputs 2102 can be an annotation channel 2164 generated by the annotation generator 2472 shown in FIG. 24. In one implementation, annotation channel 2154 can be generated based on molecule processing annotation 2402. In another implementation, annotation channel 2154 can be generated based on region annotation 2412. In yet another implementation, annotation channel 2154 can be generated based on site annotation 2422. In yet another embodiment, annotation channel 2154 can be generated based on amino acid modification annotation 2432. In yet another implementation, annotation channel 2154 can be generated based on secondary structure annotation 2442. In yet another implementation, annotation channel 2154 can be generated based on experiment information annotation 2452.

Another of the inputs 2102 may be a structural confidence channel 2174 generated by a structural confidence generator 2572 shown in FIG. 25. In one implementation, structural confidence 2174 may be generated based on global model quality estimation (GMQE) 2502. In another implementation, structural confidence 2174 can be generated based on qualitative model energy analysis (QMEAN) scores 2512. In yet another implementation, structural reliability 2174 can be generated based on temperature factor 2522. In yet another implementation, structural confidence 2174 can be generated based on template modeling score 2542. Examples of template modeling scores 2542 include minimum template modeling score 2542a, average template modeling score 2542b, and maximum template modeling score 2542c.

Those skilled in the art will appreciate that any permutation and combination of input channels can be coupled to the input for processing through the pathogenicity classifier 2108 for pathogenicity determination 2106 of the target variant. In some implementations, only a subset of input channels may be concatenated. Input channels can be concatenated in any order. In one implementation, the input channels may be concatenated into a single tensor by a tensor generator (input encoder) 2110. This single tensor can then be provided as an input to a pathogenicity classifier 2108 for pathogenicity determination 2106 of the target variant.

In one implementation, pathogenicity classifier 2108 uses a convolutional neural network (CNN) with multiple convolutional layers. In another embodiment, the pathogenicity classifier 2108 uses recurrent neural networks (RNNs), such as long short-term memory networks (LSTMs), bidirectional LSTMs (Bi-LSTMs), and gated recurrent units (GRUs). do. In yet another embodiment, pathogenicity classifier 2108 uses both CNNs and RNNs. In yet another implementation, pathogenicity classifier 2108 uses a graph convolutional neural network that models dependencies in graph structured data. In yet another embodiment, pathogenicity classifier 2108 uses a variational autoencoder (VAE). In yet another embodiment, pathogenicity classifier 2108 uses a generative adversarial network (GAN). In yet another implementation, pathogenicity classifier 2108 may be a self-attention-based language model, such as those implemented by Transformer and BERT, for example.

In still other embodiments, the pathogenicity classifier 2108 includes 1D convolution, 2D convolution, 3D convolution, 4D convolution, 5D convolution, dilated or dilated convolution, transposed convolution, depth separable convolution, pointwise convolution, 1x1 Convolutions, group convolutions, flattened convolutions, spatial and cross-channel convolutions, shuffled grouped convolutions, spatially separable convolutions, and deconvolutions can be used. The pathogenicity classifier 2108 is one of a logistic regression/log loss, a multiclass cross entropy/softmax loss, a binary cross entropy loss, a mean squared error loss, an L1 loss, an L2 loss, a smoothed L1 loss, and a Huber loss. The above loss functions can be used. It includes TFRecord, compression encoding (e.g. PNG), sharpening, parallel calls to map transforms, batching, prefetching, model parallelism, data parallelism, and synchronous/asynchronous stochastic gradient descent (SGD). Any parallelism, efficiency, and compression scheme can be used. It includes upsampling layers, downsampling layers, recurrent connections, gates and gated memory units (such as LSTM or GRU), residual blocks, residual connections, highway connections, skip connections, peephole connections, activation functions (e.g. , regularized linear unit (ReLU), Leaky ReLU, exponential linear unit (ELU), nonlinear transformation functions such as sigmoid and hyperbolic tangent function (tanh), batch normalization layer, regularization layer, dropout, pooling layer (e.g., max or average pooling), a global average pooling layer, an attention mechanism, and a Gaussian error linear unit.

Pathogenicity classifier 2108 is trained using a backpropagation-based gradient update technique. Exemplary gradient descent techniques that the pathogenicity classifier 2108 may use to learn include stochastic gradient descent, batch gradient descent, and mini-batch gradient descent. Some examples of gradient descent optimization algorithms that the pathogenicity classifier 2108 can use to learn include Momentum, Nestorv accelerated gradient, Adagrad, Adadelta, RMSprop, Adam, AdaMax, Nadam, and AMSGrad. It will be done. In other implementations, the pathogenicity classifier 2108 can be trained by unsupervised learning, semi-supervised learning, self-learning, reinforcement learning, multi-task learning, multimodal learning, transfer learning, knowledge distillation, etc.

FIG. 26 illustrates an example processing architecture 2600 for the pathogenicity classifier 2108, according to one implementation of the disclosed technology. Processing architecture 2600 includes a cascade of processing modules 2606, 2610, 2614, 2618, 2622, 2626, 2630, 2634, 2638, and 2642, each of which performs 1D convolution (1x1x1 CONV), 3D convolution ( 3x3x3 CONV), ReLU nonlinearity, and batch normalization (BN). Other examples of processing modules include a fully connected (FC) layer, a dropout layer, a flattening layer, and a final software that generates exponentially normalized scores for target variants belonging to benign and pathogenic classes. Including max tier. In FIG. 26, "64" indicates the number of convolution filters applied by a particular processing module. In FIG. 26, the size of input voxel 2602 is 15×15×15×8. FIG. 26 also shows the volumetric dimensions of each of intermediate inputs 2604, 2608, 2612, 2616, 2620, 2624, 2628, 2632, 2636, and 2640 generated by processing architecture 2600.

FIG. 27 shows an example processing architecture 2700 for the pathogenicity classifier 2108, according to one implementation of the disclosed technology. Processing architecture 2700 includes processing modules 2708, 2714, 2720, 2726, 2732, 2738, 2744, 2750, such as 1D convolution (CONV 1D), 3D convolution (CONV 3D), ReLU nonlinearity, and batch normalization (BN). Contains a cascade of 2756, 2762, 2768, 2774, and 2780. Other examples of processing modules are fully connected (dense) layers, dropout layers, flattening layers, and final software that generates exponentially normalized scores for target variants belonging to benign and pathogenic classes. Including max tier. In FIG. 27, "64" and "32" indicate the number of convolution filters applied by a particular processing module. In Figure 27, the size of input voxels 2704 provided by input layer 2702 is 7x7x7x108. FIG. 27 also shows intermediate inputs 2710, 2716, 2722, 2728, 2734, 2740, 2746, 2752, 2758, 2764, 2770, 2776, and 2782 generated by processing architecture 2700 and resulting intermediate outputs 2706, 2712. , 2718, 2724, 2730, 2736, 2742, 2748, 2754, 2760, 2766, 2772, 2778, and 2784.

Those skilled in the art will be able to develop other current and future artificial intelligence, machine learning, and deep learning models, datasets, and learning techniques based on the disclosed mutant pathogenicity without departing from the spirit of the disclosed technology. It will be appreciated that it can be incorporated into a classifier.

Performance Results as Objective Indicators of Inventiveness and Nonobviousness The variant pathogenicity classifier disclosed herein makes pathogenicity predictions based on 3D protein structures and is referred to as "PrimateAI 3D.""PrimateAI" is a co-owned, previously disclosed variant pathogenicity classifier that makes pathogenicity predictions based on protein sequence. Further details about PrimateAI can be found in co-owned U.S. patent applications 16/160,903, 16/160,986, 16/160,968, and 16/407,149, as well as Sundaram, L. et al. Predicting the clinical impact of human mutations with deep neural networks. Nat. Genet. 50, 1161-1170 (2018).

Figures 28, 29, 30, and 31 use PrimateAI as a benchmark model to demonstrate the classification superiority of PrimateAI 3D over PrimateAI. The performance results in Figures 28, 29, 30, and 31 are generated based on the classification task of accurately distinguishing benign variants from pathogenic variants across multiple validation sets. PrimateAI 3D is trained on a training set that is different from the multiple validation sets. PrimateAI 3D is trained on common human variants and primate-derived variants used as benign datasets, while simulated variants based on trinucleotide context are used as unlabeled or pseudo-pathogenic datasets.

Emerging Developmental Delay Disorders (Emerging DDD) is an example of a validation set used to compare the classification accuracy of Primate AI 3D to Primate AI. The new DDD validation set labels variants from individuals with DDD as pathogenic and the same variants from healthy relatives of individuals with DDD as benign. A similar labeling scheme is used with the autism spectrum disorder (ASD) validation set shown in FIG.

BRCA1 is another example of a validation set used to compare the classification accuracy of Primate AI 3D to Primate AI. The BRCA1 validation set labels synthetically generated reference amino acid sequences simulating the protein of the BRCA1 gene as benign variants and labels synthetically modified allelic amino acid sequences simulating the protein of the BRCA1 gene as pathogenic. Label as mutant. A similar labeling scheme is used with different validation sets of the TP53 gene, the TP53S3 gene and its variants, and the other genes and their variants shown in Figure 31.

FIG. 28 identifies the performance of the benchmark PrimateAI model with blue horizontal bars and the performance of the disclosed PrimateAI 3D model with orange horizontal bars. The green horizontal bar indicates the pathogenicity prediction derived by combining the respective pathogenicity predictions of the disclosed PrimateAI 3D model and the benchmark PrimateAI model. In the legend, "ens10" indicates an ensemble of 10 PrimateAI 3D models, each trained on a different seed training dataset and randomly initialized with different weights and biases. Also, "7x7x7x2" indicates the size of the voxel grid used to encode the input channel during training of the ensemble of 10 PrimateAI 3D models. For a given variant, an ensemble of 10 PrimateAI 3D models each generates 10 pathogenicity predictions, which are then combined (e.g., by averaging) to form the final prediction for a given variant. generate predictive pathogenicity predictions. This logic applies equally to ensembles of different group sizes.

Also in FIG. 28, the y-axis has different validation sets and the x-axis has p-values. A larger p-value, ie, a longer horizontal bar, indicates a higher accuracy in distinguishing benign from pathogenic variants. As evidenced by the p-values in Figure 28, PrimateAI 3D outperforms PrimateAI across most of the validation set (the only exception being the tp53s3_A549 validation set). That is, PrimateAI 3D's orange horizontal bar is consistently longer than PrimateAI's blue horizontal bar.

Also in FIG. 28, the "average" category along the y-axis calculates the average of the p-values determined for each of the validation sets. Even in the average category, PrimateAI 3D outperforms PrimateAI.

In Figure 29, PrimateAI is represented by the blue horizontal bar, the ensemble of 20 PrimateAI 3D models trained on a voxel grid of size 3x3x3 is represented by the red horizontal bar, the ensemble of 10 PrimateAI 3D models trained on a voxel grid of size 7x7x7 is represented by the purple horizontal bar, the ensemble of 20 PrimateAI 3D models trained on a voxel grid of size 7x7x7 is represented by the brown horizontal bar, and the ensemble of 20 PrimateAI 3D models trained on a voxel grid of size 17x17x17 is represented by the purple horizontal bar.

Also in FIG. 29, the y-axis has different validation sets and the x-axis has p-values. As before, a larger p-value, i.e. a longer horizontal bar, indicates a higher accuracy in distinguishing benign from pathogenic variants. As demonstrated by the p-values in Figure 20, the different configurations of PrimateAI 3D outperform PrimateAI across the majority of the validation set. That is, PrimateAI 3D's red, purple, brown, and pink horizontal bars are mostly longer than PrimateAI's blue horizontal bars.

Also in FIG. 29, the "average" category along the y-axis calculates the average of the p-values determined for each of the validation sets. Even in the average category, different configurations of PrimateAI 3D outperform PrimateAI.

In Figure 30, the red vertical bar represents PrimateAI and the cyan vertical bar represents PrimateAI 3D. In Figure 30, the y-axis has p-values and the x-axis has different validation sets. In FIG. 30, without exception, PrimateAI 3D consistently outperforms PrimateAI across all of the validation sets. That is, the PrimateAI 3D cyan vertical bar is always longer than the PrimateAI red vertical bar.

FIG. 31 identifies the performance of the benchmark PrimateAI model with blue vertical bars and the performance of the disclosed PrimateAI 3D model with orange vertical bars. The green vertical bar indicates the pathogenicity prediction derived by combining the respective pathogenicity predictions of the disclosed PrimateAI 3D model and the benchmark PrimateAI model. In Figure 31, the y-axis has p-values and the x-axis has different validation sets.

As demonstrated by the p-values in Figure 31, PrimateAI 3D outperforms PrimateAI across the majority of the validation set (the only exception being the tp53s3_A549_p53NULL_Nutlin-3 validation set). That is, the orange vertical bars in PrimateAI 3D are consistently longer than the blue vertical bars in PrimateAI.

Also in FIG. 31, a separate "average" chart calculates the average of the p-values determined for each of the validation sets. PrimateAI 3D also outperforms PrimateAI in the average chart.

Average statistics can be biased by outliers. To address this, a separate "method rank" chart is also shown in FIG. The higher the rank, the worse the classification accuracy. Similarly in the method rank chart, PrimateAI 3D outperforms PrimateAI by having more counts of lower ranks 1 and 2 versus Primate AI which has all 3s.

It is also clear in Figures 28-31 that superior classification accuracy is obtained by combining PrimateAI 3D with PrimateAI. That is, a protein can be fed to PrimateAI 3D as an amino acid sequence to generate a first output, the same protein can be fed to PrimateAI 3D as a 3D voxelized protein structure to generate a second output, and the same protein can be fed to PrimateAI 3D as a 3D voxelized protein structure to generate a second output. The first and second outputs can be analyzed in combination or as a whole to generate a final pathogenicity prediction of the variant experienced by the protein.

Efficient Voxelization FIG. 32 is a flowchart illustrating an efficient voxelization process 3200 that efficiently identifies the nearest atoms for each voxel.

Here, the distance channel will be explained again. As mentioned above, the reference amino acid sequence 202 can contain different types of atoms, such as alpha carbon atoms, beta carbon atoms, oxygen atoms, nitrogen atoms, hydrogen atoms, etc. Thus, as discussed above, the distance channels may be arranged by the nearest alpha carbon atom, the nearest beta carbon atom, the nearest oxygen atom, the nearest nitrogen atom, the nearest hydrogen atom, etc. For example, in FIG. 6, each of the nine voxels 514 has a distance channel every 21 amino acids for the nearest alpha carbon atom. 6 also has a distance channel every 21 amino acids for the nearest beta carbon atom for each of the nine voxels 514, regardless of atom type and amino acid type, and for each of the nine voxels 514. It can be further extended to also have the nearest common atom distance channel for the nearest atom. In this way, each of the nine voxels 514 can have 43 distance channels.

We now turn to the number of distance calculations required to identify the nearest atom per voxel for inclusion in the distance channel. Consider the example of FIG. 3, which shows a total of 828 alpha carbon atoms distributed across 21 amino acid categories. To calculate the distance channels 602-642 per amino acid in FIG. 6, i.e. to determine the 189 distance values, the distance from each of the nine voxels 514 to each of the 828 alpha carbon atoms is measured, resulting in 9 ^* 828=7,452 distance calculations. For a 3D case of 27 voxels, this results in 27 ^* 828=22,356 distance calculations. If the 828 beta carbon atoms are also included, this number increases to 27 ^* 1656=44,712 distance calculations.

This means that the runtime complexity of identifying the nearest atom per voxel for single protein voxelization is O(number of atoms ^* number of voxels), as illustrated by FIG. 35A. Furthermore, when distance channels are computed over various attributes (e.g., different features or channels per voxel, such as annotation channels and structural confidence channels), the runtime complexity of single protein voxelization is ^* number of voxels ^* number of attributes).

As a result, distance calculations can be the most computationally intensive part of the voxelization process, taking valuable computational resources away from important runtime tasks such as model learning and model inference. For example, consider the case of model training using a training dataset of 7,000 proteins. Generating distance channels for multiple voxels across multiple amino acids, atoms, and attributes can involve over 100 voxelizations per protein, resulting in approximately 800,000 voxelizations in a single learning iteration (epoch) . A training run of 20-40 epochs with rotation of atomic coordinates in each epoch can result in as many as 32 million voxels.

In addition to the high computational cost, the size of the data for 32 million voxelization is too large to fit into main memory (e.g., over 20 TB for a 15 x 15 x 15 voxel grid). Considering parameter optimization and performing iterative learning for ensemble learning, the memory footprint of the voxelization process is too large to be stored on disk, making the voxelization process part of model training and requiring a pre-computation step. I don't do it.

The disclosed technique provides an efficient voxelization process that achieves up to approximately 100x speedup for a runtime complexity of O (number of atoms ^* number of voxels). The disclosed efficient voxelization process reduces the runtime complexity to O (number of atoms) for single protein voxelization. For different features or channels for each voxel, the disclosed efficient voxelization process reduces the runtime complexity for single protein voxelization to O(number of atoms ^* number of attributes). As a result, the voxelization process becomes as fast as model training, shifting the computational bottleneck from voxelization back to computing neural network weights on processors such as GPUs, ASICs, TPUs, FPGAs, and CGRAs. .

In some embodiments of the disclosed efficient voxelization process involving large voxel grids, the runtime complexity for single protein voxelization is O (number of atoms) for different features or channels per voxel. + voxels) and O (number of atoms ^* number of attributes+voxels). "+Voxel" complexity is defined when the number of atoms is very small compared to the number of voxels, for example, when there is one atom in a 100x100x100 voxel grid (i.e. 1 million voxels per atom). voxels). In such a scenario, the runtime is dominated by a huge number of voxel overheads, e.g. to allocate memory for 1 million voxels, to initialize 1 million voxels to 0, etc.

Details of the disclosed efficient voxelization process will now be described. 32A, 32B, 33, 34, and 35B will be described in parallel.

Starting from FIG. 32A, in step 3202, each atom (e.g., each of the 828 alpha carbon atoms and each of the 828 beta carbon atoms) is divided into one of the voxels containing the atom (e.g., 9 voxels 514). one of them). The term "contains" refers to the 3D atomic coordinates of atoms located within the voxel. Voxels containing atoms are also referred to herein as "atom-containing voxels."

Figures 32B and 33 illustrate how voxels containing particular atoms are selected. Figure 33 uses 2D atomic coordinates to represent 3D atomic coordinates. Note that the voxel grid 522 is equally spaced with each of the voxels 514 having the same step size (eg, 1 angstrom (Å) or 2 Å).

Also in FIG. 33, voxel grid 522 has magenta indices [0,1,2] along a first dimension (e.g., the x-axis) and along a second dimension (e.g., the y-axis). and has cyan index [0, 1, 2]. Also, in FIG. 33, each voxel 514 in the voxel grid 522 is represented by a green voxel index [voxel 0, voxel 1, ..., voxel 8] and a black voxel center index [(1, 1), (1 , 2), ..., (3, 3)].

Further, in FIG. 33, the center coordinates of the voxel center along the first dimension, that is, the first dimension voxel coordinates are specified in orange. Further, in FIG. 33, the center coordinates of the voxel center along the second dimension, that is, the second dimension voxel coordinates are specified in red.

First, in step 3202a (step 1 in FIG. 33), the 3D atomic coordinates (1.7456, 2.14323) of a specific atom are quantized, and the quantized 3D atomic coordinates (1.7, 2.1) are generated. Quantization can be achieved by rounding or truncation of bits.

Then, in step 3202b (step 2 of FIG. 33), the voxel coordinates (or voxel center or voxel center coordinates) of voxel 514 are assigned to dimensionally quantized 3D atomic coordinates. For the first dimension, the quantized atomic coordinate 1.7 covers the first dimension voxel coordinates ranging from 1 to 2 and is centered at 1.5 in the first dimension, so the voxel Assigned to 1. Note that voxel 1 has index 1 along the first dimension, as opposed to index 0 along the second dimension.

For the second dimension, starting with voxel 1, voxel grid 522 is traversed along the second dimension. This assigns a quantized atomic coordinate of 2.5 to voxel 7 because voxel 7 covers the range of second-dimensional voxel coordinates from 2 to 3 and is centered at 2.5 in the second dimension. Note that voxel 7 has index 2 along the second dimension as opposed to having index 1 along the first dimension.

Next, in step 3202c (step 3 of FIG. 33), the dimension index corresponding to the assigned voxel coordinates is selected. That is, for voxel 1, index 1 is selected along the first dimension, and for voxel 7, index 2 is selected along the second dimension. Those skilled in the art will appreciate that the above steps can be performed similarly for the third dimension to select a dimension index along the third dimension.

Next, in step 3202d (step 4 of FIG. 33), a cumulative sum is generated based on position-wise weighting of the selected dimension indices by a power of the base number. The general concept behind positional numbering systems is that numbers are represented by increasing powers of a base (or base), e.g., a binary number is base 2, a ternary number is base 3 , octal numbers are base 8, and hexadecimal numbers are base 16. This is often referred to as a weighted numbering system because each position is weighted by a power of the base number. The set of valid numbers for a position numbering system is equal in size to the radix of that system. For example, the decimal system has 10 digits from 0 to 9, and the ternary system has 3 digits, 0, 1, and 2. The maximum significant number in a radix system is one less than the radix (thus, 8 is not a significant number in a radix less than 9). Any decimal integer can be represented exactly in any other integer base system, and vice versa.

Returning to the example of Figure 33, the selected dimension indices 1 and 2 are reduced to a single integer by multiplying them position by position by their respective powers of base 3 and summing the results of the position by position multiplications. converted. Radix 3 is chosen here because 3D atomic coordinates have three dimensions (although FIG. 33 only shows 2D atomic coordinates along two dimensions for simplicity).

Since index 2 is located at the rightmost bit, i.e., the least significant bit, we raise 3 to the power of 0 to get 2. Since index 1 is located at the second bit from the right (i.e., the second least significant bit), we raise 3 to the power of 1 to get 3. This results in a running sum of 5.

Next, in step 3202e (step 5 of FIG. 33), the voxel index of the voxel containing the particular atom is selected based on the cumulative sum. That is, the cumulative sum is interpreted as the voxel index of the voxel containing the particular atom.

In step 3212, after each atom is associated with the atom-containing voxel, each atom is further associated with one or more voxels in the vicinity of the atom-containing voxel, also referred to herein as "neighborhood voxels." Neighboring voxels can be selected based on being within a predetermined radius (eg, 5 angstroms (Å)) of atom-containing voxels. In other implementations, neighboring voxels can be selected based on their sequential adjacency to atom-containing voxels (eg, top, bottom, right, left neighboring voxels). The resulting associations that associate each atom with the atom-containing voxel and neighboring voxels are encoded in an atom-to-voxel mapping 3402, also referred to herein as an element-to-cell mapping. In one example, the first alpha carbon atom is associated with a first subset of voxels 3404 that includes atom-containing voxels and neighboring voxels of the first alpha carbon atom. In another example, the second alpha carbon atom is associated with a second subset of voxels 3406 that includes atom-containing voxels and neighboring voxels of the second alpha carbon atom.

Note that no distance calculations are performed to determine atom-containing voxels and neighboring voxels. Atom-containing voxels are selected by a spatial arrangement of voxels that allows assigning quantized 3D atomic coordinates (without using distance calculations) to corresponding regularly spaced voxel centers in the voxel grid. Ru. Neighboring voxels are also selected by being spatially adjacent to atom-containing voxels in the voxel grid (again, without using distance calculations).

In step 3222, each voxel is mapped to the atoms associated in steps 3202 and 3212. In one implementation, this mapping is encoded in a voxel-to-atom mapping 3412, which is an atom-to-voxel mapping (e.g., by applying a voxel-based sort key to the atom-to-voxel mapping 3402). Generated based on mapping 3402. Voxel-to-atom mapping 3412 is also referred to herein as "cell-to-element mapping." In one example, in steps 3202 and 3212, the first voxel is mapped to a first subset of alpha carbon atoms 3414 that includes alpha carbon atoms associated with the first voxel. In another example, in steps 3202 and 3212, the second voxel is mapped to a second subset of alpha carbon atoms 3416 that includes alpha carbon atoms associated with the second voxel.

In step 3232, for each voxel, the distance between the voxel and the atom mapped to the voxel in step 3222 is calculated. Step 3232 has a runtime complexity of O (number of atoms) because the distance to a particular atom is measured only once from each voxel to which a particular atom is uniquely mapped in voxel-to-atom mapping 3412. has. This is the case if neighboring voxels are not considered. If there are no neighbors, the constant coefficient implied by the big-O notation is 1. With neighborhoods, the big-O notation is equal to the number of neighborhoods + 1 since the number of neighborhoods is constant for each voxel, so the runtime complexity remains O (number of atoms). In contrast, in FIG. 35A, the distance to a particular atom is measured in duplicate by the number of voxels (eg, 27 distances for a particular atom due to 27 voxels).

In FIG. 35B, based on the voxel-to-atom mapping 3412, each voxel is mapped to a respective subset of 828 atoms (not including distance calculations to neighboring voxels), as indicated by the respective ellipses for each voxel. ). Each subset is largely non-overlapping, with some exceptions. There is slight overlap due to some instances where multiple atoms are mapped to the same voxel, as shown by the prime "'" and the yellow overlap between ellipses in FIG. 35B. This minimal overlap has an additive effect on the runtime complexity of O (number of atoms) and no multiplicative effect. This overlap is the result of considering neighboring voxels after determining the voxel containing the atom. Without neighboring voxels, an atom is associated with only one voxel, so there can be no overlap. However, considering neighborhoods, each neighborhood can potentially be associated with the same atom (unless there are other atoms of the same amino acid that are closer).

At step 3242, for each voxel, the closest atom to the voxel is identified based on the distance calculated at step 3232. In one implementation, this identification is encoded in a voxel-to-nearest atom mapping 3422, also referred to herein as a "cell-to-nearest element mapping." In one example, the first voxel is mapped to the second alpha carbon atom as its nearest alpha carbon atom 3424. In another example, the second voxel is mapped to the 31st alpha carbon atom as its nearest alpha carbon atom 3426.

Furthermore, since the voxel-wise distances are calculated using the techniques described above, the atom type and amino acid type classification of the atoms and the corresponding distance values are stored to generate the classified distance channel.

Once the distances to the nearest atoms are determined using the techniques described above, these distances can be encoded in a distance channel for voxelization and subsequent processing by the pathogenicity classifier 2108. can.

Computer System Figure 36 illustrates an exemplary computer system 3600 that can be used to implement the disclosed techniques. Computer system 3600 includes at least one central processing unit (CPU) 3672 that communicates with a number of peripheral devices via a bus subsystem 3655. These peripheral devices can include, for example, a storage subsystem 3610 including memory devices and a file storage subsystem 3636, user interface input devices 3638, user interface output devices 3676, and a network interface subsystem 3674. The input and output devices enable user interaction with computer system 3600. Network interface subsystem 3674 provides an interface to external networks, including interfaces to corresponding interface devices in other computer systems.

In one implementation, pathogenicity classifier 2108 is communicatively linked to storage subsystem 3610 and user interface input device 3638.

User interface input devices 3638 include pointing devices such as keyboards, mice, trackballs, touch pads, or graphics tablets, scanners, touch screens integrated into displays, audio input devices such as voice recognition systems and microphones, and other type of input device. In general, use of the term "input device" is intended to include all possible types of devices and methods for inputting information into computer system 3600.

User interface output device 3676 may include a display subsystem, a printer, a fax machine, or a non-visual display such as an audio output device. The display subsystem includes a flat panel device such as an LED display, a cathode ray tube (CRT), a liquid crystal display (LCD), a projection device, or some other mechanism for creating a visible image. be able to. The display subsystem may also provide non-visual displays such as audio output devices. In general, use of the term "output device" is intended to include all possible types of devices and methods for outputting information from computer system 3600 to a user or another machine or computer system.

Storage subsystem 3610 stores programming and data constructs that provide functionality for some or all of the modules and methods described herein. These software modules are typically executed by processor 3678.

Processor 3678 may include a graphics processing unit (GPU), field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), and/or coarse-grained It can be a coarse-grained reconfigurable architecture (CGRA). Processor 3678 may be hosted by a deep learning cloud platform such as Google Cloud Platform(TM), Xilinx(TM), and Cirrascale(TM). Examples of processors 3678 include Google's Tensor Processing Unit (TPU)(TM), rackmount solutions such as the GX4 Rackmount Series(TM), GX36 Rackmount Series(TM), NVIDIA DGX-1(TM), Micro soft' Stratix V FPGA(TM), Graphcore's Intelligent Processor Unit (IPU)(TM), Qualcomm's Zeroth Platform(TM) with Snapdragon processors(TM), NVIDIA's Volta(TM) ), NVIDIA's DRIVE PX (trademark), NVIDIA's JETSON Lamb with TX1/TX2 MODULE(TM), Intel's Nirvana(TM), Movidius VPU(TM), Fujitsu DPI(TM), ARM's DynamicIQ(TM), IBM TrueNorth(TM), Testa VI 00s(TM) da GPU Server, and others.

The memory subsystem 3622 used in the storage subsystem 3610 includes a main random access memory (RAM) 3632 for storing instructions and data during program execution, and a read-only memory (RAM) 3632 for storing fixed instructions. read only memory (ROM) 3634. File storage subsystem 3636 may provide persistent storage for program and data files and may include a hard disk drive, associated removable media, CD-ROM drive, optical drive, or removable media cartridge. can include. Modules that implement the functionality of particular embodiments may be stored by file storage subsystem 3636 within storage subsystem 3610 or in other machines accessible by a processor.

Bus subsystem 3655 provides a mechanism for allowing the various components and subsystems of computer system 3600 to communicate with each other as intended. Although bus subsystem 3655 is shown schematically as a single bus, alternative implementations of the bus subsystem may use multiple buses.

Computer system 3600 itself may include a personal computer, a portable computer, a workstation, a computer terminal, a network computer, a television, a mainframe, a server farm, a loosely distributed set of loosely networked computers, or any other data processing system. or may be of various types including user devices. Due to the ever-changing nature of computers and networks, the description of computer system 3600 shown in FIG. 36 is intended only as a specific example for purposes of illustrating a preferred embodiment of the present invention. Many other configurations of computer system 3600 may have more or fewer components than the computer system shown in FIG. 36.

Specific embodiment 1
The following embodiments can be implemented as a system, method, or article of manufacture. One or more features of the embodiments can be combined with the base embodiments. It is taught that embodiments that are not mutually exclusive are combinable. One or more features of an embodiment may be combined with other embodiments. This disclosure regularly informs users of these options. Omission from some embodiments of enumerations that repeat these options should not be construed as limiting the combinations taught in the preceding sections. These descriptions are incorporated herein by reference to each of the following implementations.

Although the disclosed techniques use 3D data as input, other implementations may similarly use 1D data, 2D data (e.g., pixel and 2D atomic coordinates), 4D data, 5D data, etc. .

In some embodiments, the system includes a memory that stores per amino acid distance channels for a plurality of amino acids in the protein. Each of the per amino acid distance channels has a per voxel distance value for a voxel in the plurality of voxels. The per voxel distance value specifies a distance from a corresponding voxel in the plurality of voxels to an atom of a corresponding amino acid in the plurality of amino acids. The system further includes a pathogenicity determination engine configured to process a tensor that includes the per amino acid distance channels and alternative alleles of the protein expressed by the variant. The pathogenicity determination engine can also be configured to determine a pathogenicity of the variant based at least in part on the tensor.

In some embodiments, the system further includes a distance channel generator that centers the voxel grid of voxels on the alpha carbon atom of each residue of the amino acid. A distance channel generator can center a voxel grid on the alpha carbon atom of a particular amino acid residue that is located in a variant amino acid in a protein.

The system may be configured to encode the orientation of the amino acid and the position of the particular amino acid in the tensor by multiplying the per-voxel distance value for the amino acid preceding the particular amino acid by a directionality parameter. The distance may be the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom of the corresponding amino acid. In some implementations, the nearest atom distance may be a Euclidean distance. The nearest atom distance may be normalized by dividing the Euclidean distance by the maximum nearest atom distance. The amino acid may have an alpha carbon atom, and in some implementations, the distance may be the nearest alpha carbon atom distance from the corresponding voxel center to the nearest alpha carbon atom of the corresponding amino acid. The amino acid may have a beta carbon atom, and in some implementations, the distance may be the nearest beta carbon atom distance from the corresponding voxel center to the nearest beta carbon atom of the corresponding amino acid. The amino acid may have a backbone atom, and in some implementations, the distance may be the nearest backbone atom distance from the corresponding voxel center to the nearest backbone atom of the corresponding amino acid. The amino acids have side chain atoms, and in some embodiments, the distance can be the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding amino acid.

The system can be further configured to encode, in the tensor, a nearest atom channel that specifies the distance from each voxel to the nearest atom. The closest atom can be selected without regard to the amino acid and the atomic element of the amino acid. In some implementations, the distance is a Euclidean distance. The distance can be normalized by dividing the Euclidean distance by the maximum distance. Amino acids can include non-standard amino acids. The tensor can include an absent atom channel that specifies atoms not found within a predetermined radius of the voxel center, and the absent atom channel can be one-hot encoded. In some implementations, the tensor can further include one-hot encoding of alternative alleles encoded voxel-wise in each of the per-amino acid distance channels. The tensor can further include reference alleles of the protein. In some implementations, the tensor can further include one-hot encoding of the reference allele encoded voxel-wise in each of the per-amino acid distance channels. The tensor can further include an evolutionary profile that specifies the level of conservation of amino acids across multiple species.

The system further selects, for each voxel, the closest atom across amino acids and atom categories, selects a pan-amino acid conserved frequency sequence for the amino acid residue containing the closest atom, and converts the pan-amino acid conserved frequency sequence into an evolutionary profile. One can include an evolutionary profile generator that is made available. A pan-amino acid conserved frequency sequence can be constructed for specific positions of residues as observed in multiple species. A pan-amino acid conservation frequency array can specify whether there is a conserved frequency that is missing for a particular amino acid. In some embodiments, the evolutionary profile generator may select, for each of the voxels, the respective closest atom in each of the amino acids, and each residue of the amino acid that contains the closest atom. The conservation frequency for each amino acid can be selected, and the conservation frequency for each amino acid can be made available as one of the evolutionary profiles. Conserved frequencies for each amino acid can be constructed for particular positions of residues as observed in multiple species. The conservation frequency for each amino acid can specify whether there is a deletion frequency for a particular amino acid.

In some embodiments of the system, the tensor can further include an annotation channel for amino acids. Annotation channels can be one-hot encoded in tensors. Annotation channels can be molecular processing annotations including initiator methionine, signal, transit peptide, propeptide, chain, and peptide. Annotation channels can be topological domains, transmembrane, intramembrane, domains, repeats, calcium binding, zinc fingers, deoxyribonucleic acid (DNA) binding, nucleotide binding, regions, coiled coils, motifs, and regional annotations including compositional biases. can. Annotation channels can be site annotations including active sites, metal binding, binding sites, and sites. Annotation channels can be amino acid modification annotations, including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks. Annotation channels can be secondary structure annotations including helices, turns, and beta strands. Annotation channels can be experimental information annotations that include mutagenesis, sequence uncertainties, sequence conflicts, non-contiguous residues, and non-terminal residues.

In some embodiments of the system, the tensor further includes an amino acid structure confidence channel that specifies the structural quality of each of the amino acids. The structural confidence channel may be a global model quality estimate (GMQE). The structural confidence channel may include a qualitative model energy analysis (QMEAN) score. A structure confidence channel can be a temperature factor that specifies the degree to which residues meet the physical constraints of their respective protein structure. The structure confidence channel can be a template structure alignment that specifies the extent to which the residues of the atoms closest to the voxel have an aligned template structure. The structure confidence channel can be a template modeling score for aligned template structures. The structural confidence channel can be the minimum of the template modeling scores, the average of the template modeling scores, and the maximum of the template modeling scores.

In some implementations, the system can further include a tensor generator that concatenates the per-amino acid distance channels for alpha carbon atoms voxel-wise with one-hot encoding of alternative alleles to generate a tensor. . A tensor generator can concatenate per-amino acid distance channels for beta carbon atoms with one-hot encoding of alternative alleles voxel-wise to generate a tensor. The tensor generator can concatenate per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, and one-hot encoding of alternative alleles voxel-wise to generate a tensor. can. The tensor generator concatenates per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, and pan-amino acid conservation frequencies voxel-wise to create a tensor. can be generated. The tensor generator concatenates per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, pan-amino acid conservation frequencies, and annotation channels on a voxel-by-voxel basis. You can generate a tensor using The tensor generator provides a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, a pan-amino acid conservation frequency, an annotation channel, and a structure confidence channel. Each voxel can be concatenated to generate a tensor. The tensor generator provides a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, and a per-amino acid conservation frequency for each of the amino acids per voxel. can be concatenated with to generate a tensor. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, a per-amino acid conservation frequency for each of the amino acids, and an annotation channel. can be concatenated voxel by voxel to generate a tensor. The tensor generator includes a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, a per-amino acid conservation frequency for each of the amino acids, an annotation channel, and structure confidence channels can be concatenated voxel by voxel to generate a tensor. The tensor generator concatenates the per-amino acid distance channel for alpha carbon atoms, the per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, and one-hot encoding of reference allele on a voxel-by-voxel basis. You can generate a tensor by doing this. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and pan-amino acid conservation frequencies. can be concatenated voxel by voxel to generate a tensor. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, pan-amino acid conservation frequencies, and annotation channels can be concatenated voxel by voxel to generate a tensor. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, pan-amino acid conservation frequencies, The annotation channel and the structural confidence channel can be concatenated voxel by voxel to generate a tensor. The tensor generator has a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, one-hot encoding of the reference allele, and a per-amino acid distance channel for each of the amino acids. The conserved frequencies for each amino acid can be concatenated voxel by voxel to generate a tensor. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and a per-amino acid distance channel for each of the amino acids. Conserved frequencies for each amino acid and annotation channels can be concatenated voxel by voxel to generate a tensor. The tensor generator has a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and a per-amino acid distance channel for each of the amino acids. The conservation frequency, annotation channel, and structure confidence channel for each amino acid can be concatenated voxel by voxel to generate a tensor.

In some embodiments, the system can further include an atomic rotation engine that rotates the atoms of the amino acids before the per-amino acid distance channels are generated. The pathogenicity decision engine can be a neural network. In certain embodiments, the pathogenicity determination engine can be a convolutional neural network. Convolutional neural networks use 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer. be able to. The 1×1×1 convolution and the 3×3×3 convolution can be three-dimensional convolutions.

In some implementations, a 1×1×1 convolutional layer may process a tensor and produce an intermediate output that is a convolved representation of the tensor. An array of 3x3x3 convolutional layers can process the intermediate output and produce a flattened output. A fully connected layer can process the flattened output and produce a denormalized output. A softmax classification layer can process the unnormalized output and produce an exponentially normalized output that identifies the likelihood that variants are pathogenic and benign. A sigmoid layer can process the unnormalized output and produce a normalized output that identifies the likelihood that a variant is pathogenic. Voxels, atoms, and distances can have three-dimensional coordinates. The tensor can have at least three dimensions, the intermediate output can have at least three dimensions, and the flattened output can have one dimension.

In some embodiments, the pathogenicity determination engine is a recurrent neural network. In other embodiments, the pathogenicity determination engine is an attention-based neural network. In yet other embodiments, the pathogenicity determination engine is a gradient boosted tree. In yet other implementations, the pathogenicity determination engine is a state vector machine.

In other embodiments, the system can include a memory that stores distance channels for each atomic category for amino acids in a protein. Amino acids can have atoms of multiple atomic categories, and atomic categories within multiple atomic categories can specify the atomic elements of the amino acid. A distance channel for each atom category can have per-voxel distance values for voxels within a plurality of voxels. The per-voxel distance value may specify the distance from a corresponding voxel within the plurality of voxels to an atom within the corresponding atom category within the plurality of atom categories. The system is configured to process a tensor including distance channels for each atomic category and alternative alleles of a protein expressed by the variant, and to determine pathogenicity of the variant based at least in part on the tensor. A sex determination engine may further be included.

The system can further include a distance channel generator that centers a voxel grid of voxels on a respective atom of a respective atom category within the plurality of atom categories. The distance channel generator can center the voxel grid on an alpha carbon atom of a residue of at least one variant amino acid in the protein. The distance may be the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom in the corresponding atom category. The distance of the closest atoms can be the Euclidean distance. The nearest atom distance can be normalized by dividing the Euclidean distance by the largest nearest atom distance. The distance can be the nearest atom distance from the corresponding voxel center to the nearest atom in the voxel grid, regardless of the amino acid and the atomic category of the amino acid. The distance of the closest atoms can be the Euclidean distance. The nearest atom distance can be normalized by dividing the Euclidean distance by the largest nearest atom distance.

Other implementations of the methods described in this section may include a non-transitory computer-readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation of the methods described in this section may include a system including a memory and one or more processors operable to execute the instructions stored in the memory to perform any of the methods described above.

clause set 1
1. A computer-implemented method, the method comprising:
Memorizing distance channels for each amino acid for multiple amino acids in a protein,
each of the per-amino acid distance channels having per-voxel distance values for voxels within the plurality of voxels;
storing, a per-voxel distance value specifying a distance from a corresponding voxel in the plurality of voxels to an atom of a corresponding amino acid in the plurality of amino acids;
processing a tensor containing per-amino acid distance channels and alternative alleles of the protein expressed by the mutant;
and determining pathogenicity of a variant based at least in part on the tensor.
2. 2. The computer-implemented method of clause 1, further comprising centering the voxel grid of voxels on the alpha carbon atom of each residue of the amino acid.
3. 3. The computer-implemented method of clause 2, further comprising centering the voxel grid on the alpha carbon atom of a particular amino acid residue that corresponds to at least one variant amino acid in the protein.
4. Clause 3, further comprising encoding the directionality of the amino acid and the position of the particular amino acid in the tensor by multiplying the per-voxel distance value for the amino acid preceding the particular amino acid by a directional parameter. computer implementation method.
5. 4. The computer-implemented method of clause 3, wherein the distance is the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom of the corresponding amino acid.
6. 6. The computer-implemented method of clause 5, wherein the distance between the closest atoms is the Euclidean distance.
7. 7. The computer-implemented method of clause 6, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
8. 6. The computer-implemented method of clause 5, wherein the amino acid has an alpha carbon atom and the distance is the distance of the nearest alpha carbon atom from the corresponding voxel center to the nearest alpha carbon atom of the corresponding amino acid.
9. 6. The computer-implemented method of clause 5, wherein the amino acid has a beta carbon atom and the distance is the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding amino acid.
10. 6. The computer-implemented method of clause 5, wherein the amino acid has a backbone atom and the distance is the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding amino acid.
11. 6. The computer-implemented method of clause 5, wherein the amino acid has a side chain atom and the distance is the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding amino acid.
12. The computer-implemented method of clause 3, further comprising encoding, in the tensor, a nearest atom channel specifying the distance from each voxel to the amino acid and the nearest atom selected regardless of the atomic element of the amino acid. .
13. 13. The computer-implemented method of clause 12, wherein the distance is a Euclidean distance.
14. 14. The computer-implemented method of clause 13, wherein the distance is normalized by dividing the Euclidean distance by the maximum distance.
15. 13. The computer-implemented method of clause 12, wherein the amino acids include non-standard amino acids.
16. 2. The computer-implemented method of clause 1, wherein the tensor further comprises an absent atom channel that specifies atoms not found within a predetermined radius of a voxel center, the absent atom channel being one-hot encoded.
17. 2. The computer-implemented method of clause 1, wherein the tensor further comprises one-hot encoding of alternative alleles encoded voxel-wise in each of the per-amino acid distance channels.
18. 2. The computer-implemented method of clause 1, wherein the tensor further comprises a reference allele of the protein.
19. 19. The computer-implemented method of clause 18, wherein the tensor further comprises one-hot encoding of the reference allele encoded voxel-by-voxel in each of the per-amino acid distance channels.
20. 2. The computer-implemented method of clause 1, wherein the tensor further comprises an evolutionary profile specifying the level of conservation of amino acids across multiple species.
21. For each voxel,
selecting the closest atoms across amino acids and atom categories;
selecting a pan-amino acid conserved frequency sequence for amino acid residues containing the closest atoms;
21. The computer-implemented method of clause 20, further comprising making available a pan-amino acid conserved frequency sequence as one of the evolutionary profiles.
22. 22. The computer-implemented method of clause 21, wherein a pan-amino acid conserved frequency sequence is constructed for specific positions of residues as observed in multiple species.
23. 22. The computer-implemented method of clause 21, wherein the pan-amino acid conserved frequency array specifies whether there is a missing conserved frequency for a particular amino acid.
24. For each voxel,
selecting each nearest atom in each of the amino acids;
selecting a conservation frequency for each amino acid for each residue of the amino acid containing the nearest atom;
22. The computer-implemented method of clause 21, further comprising: creating a conservation frequency for each amino acid that can be used as one of the evolutionary profiles.
25. 25. The computer-implemented method of clause 24, wherein a per-amino acid conservation frequency is constructed for a particular position of the residue as observed in multiple species.
26. 25. The computer-implemented method of clause 24, wherein the conservation frequency for each amino acid specifies whether there is a missing conservation frequency for a particular amino acid.
27. 2. The computer-implemented method of clause 1, wherein the tensor further comprises an annotation channel of amino acids, the annotation channel being one-hot encoded in the tensor.
28. 28. The computer-implemented method of clause 27, wherein the annotation channel is a molecular processing annotation that includes an initiator methionine, a signal, a transit peptide, a propeptide, a chain, and a peptide.
29. Clause where the annotation channel is a region annotation including topological domain, transmembrane, intramembrane, domain, repeat, calcium binding, zinc finger, deoxyribonucleic acid (DNA) binding, nucleotide binding, region, coiled coil, motif, and compositional bias. 28. The computer implementation method according to 27.
30. 28. The computer-implemented method of clause 27, wherein the annotation channel is a site annotation that includes an active site, a metal binding, a binding site, and a site.
31. 28. The computer-implemented method of clause 27, wherein the annotation channel is an amino acid modification annotation including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks.
32. 28. The computer-implemented method of clause 27, wherein the annotation channel is a secondary structure annotation including helices, turns, and beta strands.
33. 28. The computer-implemented method of clause 27, wherein the annotation channel is an experimental information annotation that includes mutagenesis, sequence uncertainty, sequence conflicts, non-adjacent residues, and non-terminal residues.
34. 2. The computer-implemented method of clause 1, wherein the tensor further comprises an amino acid structural confidence channel that specifies the structural quality of each of the amino acids.
35. 35. The computer-implemented method of clause 34, wherein the structural confidence channel is a global model quality estimation (GMQE).
36. 35. The computer-implemented method of clause 34, wherein the structural confidence channel includes a qualitative model energy analysis (QMEAN) score.
37. 35. The computer-implemented method of clause 34, wherein the structure confidence channel is a temperature factor that specifies the degree to which the residues meet the physical constraints of the respective protein structure.
38. 35. The computer-implemented method of clause 34, wherein the structure confidence channel is a template structure alignment that specifies the degree to which the residues of atoms closest to the voxel have aligned template structures.
39. 39. The computer-implemented method of clause 38, wherein the structure confidence channel is a template modeling score of the aligned template structure.
40. 40. The computer-implemented method of clause 39, wherein the structural confidence channels are a minimum of the template modeling scores, an average of the template modeling scores, and a maximum of the template modeling scores.
41. 2. The computer-implemented method of clause 1, further comprising concatenating the per-amino acid distance channel for the alpha carbon atom voxel-wise with one-hot encoding of alternative alleles to generate a tensor.
42. 42. The computer-implemented method of clause 41, further comprising concatenating the per-amino acid distance channel for the beta carbon atom voxel-wise with one-hot encoding of alternative alleles to generate a tensor.
43. Clause 42, further comprising concatenating the per-amino acid distance channel for the alpha carbon atom, the per-amino acid distance channel for the beta carbon atom, and the one-hot encoding of alternative alleles on a voxel-by-voxel basis to generate a tensor. Computer implementation method described.
44. Voxel-wise concatenation of per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, and pan-amino acid conserved frequency arrays to generate a tensor. 44. The computer-implemented method of clause 43, further comprising:
45. A per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, a pan-amino acid conserved frequency array, and an annotation channel are concatenated voxel-wise to create a tensor. 45. The computer-implemented method of clause 44, further comprising generating.
46. Voxel-wise concatenation of per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, pan-amino acid conserved frequency arrays, annotation channels, and structural confidence channels 46. The computer-implemented method of clause 45, further comprising generating a tensor.
47. A per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, and a per-amino acid conservation frequency for each of the amino acids are concatenated voxel by voxel. 47. The computer-implemented method of clause 46, further comprising generating a tensor.
48. Concatenation of per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, per-amino acid conservation frequencies for each of the amino acids, and annotation channels on a voxel-by-voxel basis 48. The computer-implemented method of clause 47, further comprising generating a tensor.
49. Per-amino acid distance channel for alpha carbon atoms, per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, per-amino acid conservation frequency for each of the amino acids, annotation channel, and structure confidence channel 49. The computer-implemented method of clause 48, further comprising concatenating voxel by voxel to generate a tensor.
50. Generate a tensor by voxel-wise concatenation of per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, and one-hot encoding of reference alleles. 50. The computer-implemented method of clause 49, further comprising:
51. per-amino acid distance channel for alpha carbon atoms, per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and pan-amino acid conservation frequency arrays per voxel 51. The computer-implemented method of clause 50, further comprising concatenating to generate a tensor.
52. A per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, a pan-amino acid conservation frequency array, and an annotation channel. 52. The computer-implemented method of clause 51, further comprising concatenating voxel by voxel to generate a tensor.
53. a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, a pan-amino acid conservation frequency array, an annotation channel, and 53. The computer-implemented method of clause 52, further comprising concatenating the structural confidence channels voxel by voxel to generate a tensor.
54. Per-amino acid distance channel for alpha carbon atoms, per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and per-amino acid conservation for each of the amino acids. 54. The computer-implemented method of clause 53, further comprising concatenating the frequencies voxel by voxel to generate a tensor.
55. Per-amino acid distance channel for alpha carbon atoms, per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, per-amino acid conservation frequency for each of the amino acids 55. The computer-implemented method of clause 54, further comprising concatenating the annotation channels voxel by voxel to generate a tensor.
56. Per-amino acid distance channel for alpha carbon atoms, per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, per-amino acid conservation frequency for each of the amino acids 56. The computer-implemented method of clause 55, further comprising concatenating voxel-wise the annotation channel, the annotation channel, and the structural confidence channel to generate a tensor.
57. 2. The computer-implemented method of clause 1, further comprising rotating atoms of the amino acids before the per-amino acid distance channels are generated.
58. In convolutional neural networks, use 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer The computer-implemented method of clause 1, further comprising:
59. The computer-implemented method of clause 58, wherein the 1x1x1 convolution and the 3x3x3 convolution are three-dimensional convolutions.
60. A 1x1x1 convolutional layer processes the tensor and produces an intermediate output that is a convolved representation of the tensor, and an array of 3x3x3 convolutional layers processes the intermediate output and flattens it. A fully connected layer processes the flattened output and generates an unnormalized output, a softmax classification layer processes the unnormalized output, and the variants are pathogenic and benign. 59. The computer-implemented method of clause 58, wherein the computer-implemented method generates an exponentially normalized output that identifies a likelihood.
61. 61. The computer-implemented method of clause 60, wherein the sigmoid layer processes the denormalized output and produces a normalized output that identifies the likelihood that the variant is pathogenic.
62. The computer of clause 60, wherein the voxels, atoms, and distances have three-dimensional coordinates, the tensor has at least three dimensions, the intermediate output has at least three dimensions, and the flattened output has one dimension. How to implement.
63. A computer-implemented method, the method comprising:
Storing distance channels for each atomic category for amino acids in a protein, the method comprising:
the amino acid has atoms from multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atomic category has a per-voxel distance value for a voxel within the plurality of voxels;
storing, a per-voxel distance value specifying a distance from a corresponding voxel in the plurality of voxels to an atom in the corresponding atom category in the plurality of atom categories;
processing a tensor containing distance channels for each atomic category and alternative alleles of the protein expressed by the variant;
and determining pathogenicity of a variant based at least in part on the tensor.
64. 64. The computer-implemented method of clause 63, further comprising centering the voxel grid of voxels on each atom of each atom category within the plurality of atom categories.
65. 65. The computer-implemented method of clause 64, further comprising centering the voxel grid on an alpha carbon atom of a residue of at least one variant amino acid in the protein.
66. 66. The computer-implemented method of clause 65, wherein the distance is the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom in the corresponding atom category.
67. 67. The computer-implemented method of clause 66, wherein the distance between the closest atoms is the Euclidean distance.
68. 68. The computer-implemented method of clause 67, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
69. 69. The computer-implemented method of clause 68, wherein the distance is the nearest atom distance from the corresponding voxel center to the nearest atom in the voxel grid, regardless of the amino acid and the atomic category of the amino acid.
70. 69. The computer-implemented method according to clause 69, wherein the distance between the nearest atoms is the Euclidean distance.
71. 71. The computer-implemented method of clause 70, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.

clause set 2
1. When executed on one or more processors,
storing per-amino acid distance channels for a plurality of amino acids in the protein, each per-amino acid distance channel having a per-voxel distance value for a voxel within the plurality of voxels;
A per-voxel distance value is expressed by a mutant with a per-amino acid distance channel that specifies the distance from the corresponding voxel in multiple voxels to the atom of the corresponding amino acid in multiple amino acids. processing a tensor containing alternative alleles of the protein;
one or more computer-readable media storing computer-executable instructions that configure the computer to perform operations including: determining pathogenicity of a variant based at least in part on the tensor;
2. 2. The computer-readable medium of clause 1, wherein the act further comprises centering the voxel grid of voxels on the alpha carbon atom of each residue of the amino acid.
3. 3. The computer-readable medium of clause 2, wherein the operations further include centering the voxel grid on the alpha carbon atom of a particular amino acid residue that corresponds to at least one variant amino acid in the protein.
4. Clause, wherein the operation further comprises encoding the directionality of the amino acid and the position of the particular amino acid in the tensor by multiplying the voxel-by-voxel distance value for the amino acid preceding the particular amino acid by a directionality parameter. 3. The computer readable medium according to 3.
5. 4. The computer-readable medium of clause 3, wherein the distance is the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom of the corresponding amino acid.
6. 6. The computer-readable medium of clause 5, wherein the distance between the nearest atoms is the Euclidean distance.
7. 7. The computer-readable medium of clause 6, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
8. 6. The computer-readable medium of clause 5, wherein the amino acid has an alpha carbon atom, and the distance is the distance of the nearest alpha carbon atom from the corresponding voxel center to the nearest alpha carbon atom of the corresponding amino acid.
9. 6. The computer-readable medium of clause 5, wherein the amino acid has a beta carbon atom, and the distance is the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding amino acid.
10. 6. The computer-readable medium of clause 5, wherein the amino acid has a backbone atom, and the distance is the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding amino acid.
11. 6. The computer-readable medium of clause 5, wherein the amino acid has a side chain atom, and the distance is the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding amino acid.
12. 3, wherein the operation further comprises encoding, in the tensor, a nearest atom channel specifying the distance from each voxel to the nearest atom selected without regard to the amino acid and the atomic element of the amino acid. computer readable medium.
13. 13. The computer-readable medium of clause 12, wherein the distance is a Euclidean distance.
14. 14. The computer-readable medium of clause 13, wherein the distance is normalized by dividing the Euclidean distance by the maximum distance.
15. 13. The computer-readable medium of clause 12, wherein the amino acids include non-standard amino acids.
16. 2. The computer-readable medium of clause 1, wherein the tensor further comprises an absent atom channel that specifies atoms not found within a predetermined radius of a voxel center, the absent atom channel being one-hot encoded.
17. 2. The computer-readable medium of clause 1, wherein the tensor further comprises one-hot encoding of alternative alleles encoded voxel-wise in each of the per-amino acid distance channels.
18. The computer-readable medium of clause 1, wherein the tensor further comprises a reference allele of the protein.
19. 19. The computer-readable medium of clause 18, wherein the tensor further comprises a one-hot encoding of the reference allele encoded voxel-wise in each of the per-amino acid distance channels.
20. The computer-readable medium of clause 1, wherein the tensor further comprises an evolutionary profile specifying the level of conservation of amino acids across multiple species.
21. the operation selects the closest atom across amino acids and atomic categories for each voxel;
selecting a pan-amino acid conserved frequency sequence for amino acid residues containing the closest atoms;
21. The computer-readable medium of clause 20, further comprising making available a pan-amino acid conserved frequency sequence as one of the evolutionary profiles.
22. 22. The computer-readable medium of clause 21, wherein a pan-amino acid conserved frequency sequence is constructed for specific positions of residues as observed in multiple species.
23. 22. The computer-readable medium of clause 21, wherein the pan-amino acid conserved frequency sequence specifies whether there is a missing conserved frequency for a particular amino acid.
24. the operation selects each nearest atom in each of the amino acids for each of the voxels;
selecting a conservation frequency for each amino acid for each residue of the amino acid containing the nearest atom;
22. The computer-readable medium of clause 21, further comprising: creating a conservation frequency for each amino acid that can be used as one of the evolutionary profiles.
25. 25. The computer-readable medium of clause 24, wherein a conservation frequency for each amino acid is configured for a particular position of the residue as observed in multiple species.
26. 25. The computer-readable medium of clause 24, wherein the conservation frequency for each amino acid specifies whether there is a missing conservation frequency for the particular amino acid.
27. 2. The computer-readable medium of clause 1, wherein the tensor further comprises an annotation channel of amino acids, the annotation channel being one-hot encoded in the tensor.
28. 28. The computer-readable medium of clause 27, wherein the annotation channel is a molecular processing annotation that includes an initiator methionine, a signal, a transit peptide, a propeptide, a chain, and a peptide.
29. Clause where the annotation channel is a region annotation including topological domain, transmembrane, intramembrane, domain, repeat, calcium binding, zinc finger, deoxyribonucleic acid (DNA) binding, nucleotide binding, region, coiled coil, motif, and compositional bias. 28. The computer readable medium according to 27.
30. 28. The computer-readable medium of clause 27, wherein the annotation channel is a site annotation that includes an active site, a metal binding, a binding site, and a site.
31. 28. The computer-readable medium of clause 27, wherein the annotation channel is an amino acid modification annotation including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks.
32. 28. The computer-readable medium of clause 27, wherein the annotation channel is a secondary structure annotation including helices, turns, and beta strands.
33. 28. The computer-readable medium of clause 27, wherein the annotation channel is an experimental information annotation that includes mutagenesis, sequence uncertainty, sequence conflicts, non-adjacent residues, and non-terminal residues.
34. 2. The computer-readable medium of clause 1, wherein the tensor further includes an amino acid structure confidence channel that specifies the structural quality of each of the amino acids.
35. 35. The computer-readable medium of clause 34, wherein the structural confidence channel is global model quality estimation (GMQE).
36. 35. The computer-readable medium of clause 34, wherein the structural confidence channel includes a qualitative model energy analysis (QMEAN) score.
37. 35. The computer-readable medium of clause 34, wherein the structure confidence channel is a temperature factor that specifies the degree to which the residues meet the physical constraints of their respective protein structure.
38. 35. The computer-readable medium of clause 34, wherein the structure confidence channel is a template structure alignment that specifies the degree to which the residues of atoms closest to the voxel have an aligned template structure.
39. 39. The computer-readable medium of clause 38, wherein the structure confidence channel is a template modeling score for aligned template structures.
40. 40. The computer-readable medium of clause 39, wherein the structural confidence channels are a minimum of the template modeling scores, an average of the template modeling scores, and a maximum of the template modeling scores.
41. 2. The computer-readable medium of clause 1, wherein the operations further include concatenating the per-amino acid distance channel for the alpha carbon atom voxel-wise with one-hot encoding of alternative alleles to generate a tensor.
42. 42. The computer-readable medium of clause 41, wherein the operations further include concatenating the per-amino acid distance channel for the beta carbon atom voxel-wise with one-hot encoding of alternative alleles to generate a tensor.
43. The operations further include concatenating the per-amino acid distance channel for the alpha carbon atom, the per-amino acid distance channel for the beta carbon atom, and the one-hot encoding of alternative alleles on a voxel-by-voxel basis to generate a tensor. Clause 42 Computer Readable Medium.
44. The operation consists of a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, and a voxel-wise concatenation of pan-amino acid conservation frequency arrays to create a tensor. 44. The computer-readable medium of clause 43, further comprising generating.
45. The operation combines a per-amino acid distance channel for alpha carbon atoms, a per-amino acid distance channel for beta carbon atoms, one-hot encoding of alternative alleles, a pan-amino acid conserved frequency array, and an annotation channel on a voxel-by-voxel basis. , the computer-readable medium of clause 44, further comprising generating a tensor.
46. The operations include voxel per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, pan-amino acid conserved frequency arrays, annotation channels, and structural confidence channels. 46. The computer-readable medium of clause 45, further comprising concatenating each other to generate a tensor.
47. The operations include a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, and a voxel-wise concatenation of per-amino acid conservation frequencies for each of the amino acids. 47. The computer-readable medium of clause 46, further comprising: generating a tensor.
48. The operations include a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of alternative alleles, a per-amino acid conservation frequency for each of the amino acids, and a voxel annotation channel. 48. The computer-readable medium of clause 47, further comprising concatenating each other to generate a tensor.
49. Operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, per-amino acid conservation frequencies for each of the amino acids, annotation channels, and structure. 49. The computer-readable medium of clause 48, further comprising concatenating the confidence channels voxel by voxel to generate a tensor.
50. The operation concatenates the per-amino acid distance channel for the alpha carbon atom, the per-amino acid distance channel for the beta carbon atom, one-hot encoding of the alternative allele, and one-hot encoding of the reference allele on a voxel-by-voxel basis. 50. The computer-readable medium of clause 49, further comprising generating a tensor.
51. The operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and pan-amino acid conservation frequency arrays. 51. The computer-readable medium of clause 50, further comprising concatenating voxel by voxel to generate a tensor.
52. The operations include a per-amino acid distance channel for the alpha carbon atom, a per-amino acid distance channel for the beta carbon atom, one-hot encoding of the alternative allele, one-hot encoding of the reference allele, to generate a tensor. 52. The computer-readable medium of clause 51, further comprising concatenating the pan-amino acid conserved frequency sequence and the annotation channel on a voxel-by-voxel basis.
53. The operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, pan-amino acid conservation frequency arrays, and annotations. 53. The computer-readable medium of clause 52, further comprising concatenating the channel and the structural confidence channel voxel by voxel to generate a tensor.
54. Operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, pan-amino acid conservation frequency arrays, and 54. The computer-readable medium of clause 53, further comprising concatenating the per-amino acid conservation frequencies for each of the amino acids voxel by voxel to generate a tensor.
55. Operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, pan-amino acid conserved frequency arrays, amino acids 55. The computer-readable medium of clause 54, further comprising concatenating the per-amino acid conservation frequency for each of the annotation channels voxel-wise to generate a tensor.
56. The operations include per-amino acid distance channels for alpha carbon atoms, per-amino acid distance channels for beta carbon atoms, one-hot encoding of alternative alleles, one-hot encoding of reference alleles, and per-amino acid distance channels for each of the amino acids. 56. The computer-readable medium of clause 55, further comprising concatenating the storage frequency, annotation channel, and structural confidence channel voxel by voxel to generate a tensor.
57. 2. The computer-readable medium of clause 1, wherein the operations further include rotating atoms of the amino acids before the per-amino acid distance channels are generated.
58. The operations are 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer in convolutional neural networks. The computer-readable medium of clause 1, further comprising using the computer-readable medium of clause 1.
59. The computer-readable medium of clause 58, wherein the 1x1x1 convolution and the 3x3x3 convolution are three-dimensional convolutions.
60. A 1x1x1 convolutional layer processes the tensor and produces an intermediate output that is a convolved representation of the tensor, and an array of 3x3x3 convolutional layers processes the intermediate output and flattens it. A fully connected layer processes the flattened output and generates an unnormalized output, a softmax classification layer processes the unnormalized output, and the variants are pathogenic and benign. 59. The computer-readable medium of clause 58, wherein the computer-readable medium generates an exponentially normalized output that identifies a likelihood.
61. 61. The computer-readable medium of clause 60, wherein the sigmoid layer processes the denormalized output and produces a normalized output that identifies a likelihood that the variant is pathogenic.
62. The computer of clause 60, wherein the voxels, atoms, and distances have three-dimensional coordinates, the tensor has at least three dimensions, the intermediate output has at least three dimensions, and the flattened output has one dimension. readable medium.
63. When executed on one or more processors:
Storing distance channels for each atomic category for amino acids in a protein, the method comprising:
the amino acid has atoms from multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atomic category has a per-voxel distance value for a voxel within the plurality of voxels;
storing, a per-voxel distance value specifying a distance from a corresponding voxel in the plurality of voxels to an atom in the corresponding atom category in the plurality of atom categories;
processing a tensor containing distance channels for each atomic category and alternative alleles of the protein expressed by the variant;
one or more computer-readable media storing computer-executable instructions that configure the computer to perform operations including: determining pathogenicity of a variant based at least in part on the tensor;
64. 64. The computer-readable medium of clause 63, wherein the operations further include centering a voxel grid of voxels on a respective atom of a respective atomic category within the plurality of atomic categories.
65. 65. The computer-readable medium of clause 64, wherein the act further comprises centering the voxel grid on an alpha carbon atom of a residue of at least one variant amino acid in the protein.
66. 66. The computer-readable medium of clause 65, wherein the distance is the nearest atom distance from the corresponding voxel center in the voxel grid to the nearest atom in the corresponding atom category.
67. 67. The computer-readable medium of clause 66, wherein the distance between the nearest atoms is the Euclidean distance.
68. 68. The computer-readable medium of clause 67, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
69. 69. The computer-readable medium of clause 68, wherein the distance is the nearest atom distance from the corresponding voxel center to the nearest atom in the voxel grid, regardless of the amino acid and the atomic category of the amino acid.
70. 69. The computer-readable medium of clause 69, wherein the distance between the nearest atoms is the Euclidean distance.
71. 71. The computer-readable medium of clause 70, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.

Specific embodiment 2
In some embodiments, the system accesses the three-dimensional structure of a reference amino acid sequence of the protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid. Contains a voxelizer. Each of the per-amino acid distance channels has a three-dimensional distance value for each voxel within a three-dimensional grid of voxels. The three-dimensional distance value specifies the distance from the corresponding voxel in the three-dimensional grid of voxels to the atom of the corresponding reference amino acid in the reference amino acid sequence. The system further includes an alternate allele encoder that encodes alternate allele amino acids for each voxel in the three-dimensional grid of voxels. Alternative allelic amino acids are three-dimensional representations of one-hot encoding of variant amino acids expressed by variant nucleotides. The system further includes an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels. An evolutionarily conserved sequence can be a three-dimensional representation of amino acid-specific conservation frequencies across multiple species. Amino acid-specific conservation frequencies can be selected depending on amino acid proximity to the corresponding voxel. The system further includes a convolutional neural network configured to apply a three-dimensional convolution to a tensor containing distance channels for each amino acid encoded with alternative allelic amino acids and respective evolutionarily conserved sequences. The convolutional neural network can also be configured to determine pathogenicity of a mutant nucleotide based at least in part on the tensor.

The voxelizer can center a three-dimensional grid of voxels on the alpha carbon atom of each residue of a reference amino acid within the reference amino acid sequence. The voxelizer can center a three-dimensional grid of voxels on the alpha carbon atom of a particular reference amino acid residue located at the variant amino acid residue.

In some implementations, the system determines the directionality and identification of a reference amino acid within a reference amino acid sequence by multiplying the three-dimensional distance values for reference amino acids preceding a particular reference amino acid by a directionality parameter in a tensor. can be further configured to encode the position of the reference amino acid of. The distance can be the nearest atom distance from the corresponding voxel center in a three-dimensional grid of voxels to the nearest atom of the corresponding reference amino acid. The nearest atom distance may be the Euclidean distance and can be normalized by dividing the Euclidean distance by the largest nearest atom distance.

In some embodiments, the reference amino acid can have an alpha carbon atom, and the distance can be the distance of the nearest alpha carbon atom from the corresponding voxel center to the nearest alpha carbon atom of the corresponding reference amino acid. can. In some embodiments, the reference amino acid can have a beta carbon atom, and the distance can be the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding reference amino acid. can. In some embodiments, the reference amino acid can have a backbone atom, and the distance can be the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding reference amino acid. In some embodiments, the amino acid can have a side chain atom, and the distance can be the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding reference amino acid. .

In some implementations, the system can be further configured to encode, in the tensor, a nearest atom channel that specifies the distance from each voxel to the nearest atom. The closest atom can be selected without regard to the amino acid and the atomic element of the amino acid. The distance may be a Euclidean distance and can be normalized by dividing the Euclidean distance by the maximum distance. Amino acids can include non-standard amino acids. The tensor can further include an absent atom channel that specifies atoms not found within a predetermined radius of the voxel center. Missing atomic channels can be one-hot encoded.

In some embodiments, the system can further include a reference allele encoder that encodes the reference allele amino acids voxel by voxel into each three-dimensional distance value on an amino acid position basis. The reference allele amino acid can be a one-hot encoded three-dimensional representation of the reference amino acid sequence. Amino acid-specific conservation frequencies can specify the level of conservation of each amino acid across multiple species.

In some embodiments, the evolutionary conservation encoder can select the closest atom to the corresponding voxel across the reference amino acid and atom category, and calculate the pan-amino acid conservation frequency for the residue of the reference amino acid that contains the closest atom. can be selected, and the three-dimensional representation of the pan-amino acid conservation frequency can be used as an evolutionarily conserved sequence. Pan-amino acid conservation frequencies can be constructed for specific positions of residues as observed in multiple species. The pan-amino acid conservation frequency can specify whether there are any conserved frequencies that are missing for a particular reference amino acid.

In some embodiments, the evolutionary conservation encoder can select each closest atom in each of the reference amino acids to the corresponding voxel, and each The conservation frequency for each amino acid can be selected, and the three-dimensional representation of the conservation frequency for each amino acid can be used as an evolutionarily conserved sequence. Conserved frequencies for each amino acid can be constructed for particular positions of residues as observed in multiple species. The conservation frequency for each amino acid can specify whether there is a conserved frequency that is missing for a particular reference amino acid.

In some implementations, the system can further include an annotation encoder that encodes one or more annotation channels into each three-dimensional distance value on a voxel-by-voxel basis. Annotation channels can be one-hot encoded three-dimensional representations of residue annotations and can be molecular processing annotations including initiator methionine, signal, transit peptide, propeptide, chain, and peptide. In some embodiments, annotation channels include topological domains, transmembrane, intramembrane, domains, repeats, calcium binding, zinc fingers, deoxyribonucleic acid (DNA) binding, nucleotide binding, regions, coiled coils, motifs, and compositional biases. It can be a region annotation that includes, or it can be a site annotation that includes active site, metal binding, binding site, and site. In some embodiments, annotation channels can be amino acid modified annotations, including non-canonical residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks, or helices, turns, and beta can be a secondary structure annotation that includes chains. Annotation channels can be experimental information annotations that include mutagenesis, sequence uncertainties, sequence conflicts, non-contiguous residues, and non-terminal residues.

In some implementations, the system can further include a structural confidence encoder that encodes one or more structural confidence channels into each three-dimensional distance value on a voxel-by-voxel basis. A structure confidence channel can be a three-dimensional representation of confidence scores that specify the quality of each residue structure. The structural confidence channel can be a global model quality estimate (GMQE) and can be a qualitative model energy analysis (QMEAN) score, specifying the degree to which the residues satisfy the physical constraints of the respective protein structure. can be a temperature factor that specifies the degree to which the atomic residues closest to a voxel have an aligned template structure, and can be a template modeling score for an aligned template structure. , or the minimum of the template modeling scores, the average of the template modeling scores, and the maximum of the template modeling scores.

In some embodiments, the system can further include an atom rotation engine that rotates the atoms before the per-amino acid distance channels are generated.

Convolutional neural networks use 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer. be able to. The 1×1×1 convolution and the 3×3×3 convolution can be three-dimensional convolutions. In some implementations, a 1×1×1 convolutional layer may process a tensor and produce an intermediate output that is a convolved representation of the tensor. An array of 3x3x3 convolutional layers can process the intermediate output and produce a flattened output. A fully connected layer can process the flattened output and produce a denormalized output. A softmax classification layer can process the unnormalized output and produce an exponentially normalized output that identifies the likelihood that variant nucleotides are pathogenic and benign.

In some implementations, the sigmoid layer can process the unnormalized output and produce a normalized output that identifies the likelihood that the variant nucleotide is pathogenic. A convolutional neural network can be an attention-based neural network. The tensor may include a per-amino acid distance channel further encoded with reference allele amino acids, a per-amino acid distance channel further encoded with an annotation channel, or a per-amino acid distance channel further encoded with a structural confidence channel. A distance channel can be included for each encoded amino acid.

In some embodiments, the system accesses the three-dimensional structure of a reference amino acid sequence of the protein and fits a three-dimensional grid of voxels to the atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each atom category. A voxelizer can be included. Atoms spans multiple atomic categories that specify the atomic elements of amino acids. Each of the distance channels for each atom category has a three-dimensional distance value for each voxel within a three-dimensional grid of voxels. The three-dimensional distance value specifies the distance from the corresponding voxel in the three-dimensional grid of voxels to the atom of the corresponding atom category within the plurality of atom categories. The system further includes an alternate allele encoder that encodes alternate allele amino acids for each voxel in the three-dimensional grid of voxels. Alternative allelic amino acids are three-dimensional representations of one-hot encoding of variant amino acids expressed by variant nucleotides. The system further includes an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels. An evolutionarily conserved sequence can be a three-dimensional representation of amino acid-specific conservation frequencies across multiple species. Amino acid-specific conservation frequencies can be selected depending on amino acid proximity to the corresponding voxel. The system applies three-dimensional convolution to a tensor containing distance channels for each atomic category encoded with alternative allele amino acids and their respective evolutionarily conserved sequences, and determines the pathogenicity of variant nucleotides based at least in part on the tensor. further comprising a convolutional neural network configured to determine.

In some embodiments, the system includes a voxelizer that accesses a three-dimensional structure of a reference amino acid sequence of the protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid. Each of the distance channels for each amino acid can have a three-dimensional distance value for each voxel in the three-dimensional grid of voxels. The three-dimensional distance value can specify a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence. The system further includes an alternative allele encoder that encodes an alternative allele amino acid for each voxel in the three-dimensional grid of voxels. The alternative allele amino acid is a three-dimensional representation of a one-hot encoding of a variant amino acid expressed by a variant nucleotide. The system further includes an evolutionary conservation encoder that encodes an evolutionary conserved sequence for each voxel in the three-dimensional grid of voxels. The evolutionary conserved sequence can be a three-dimensional representation of an amino acid-specific conservation frequency across multiple species. The amino acid-specific conservation frequency can be selected according to amino acid proximity to a corresponding voxel. The system further includes a tensor generator configured to generate a tensor including distance channels for each amino acid encoded with alternative allelic amino acids and their respective evolutionarily conserved sequences.

In some embodiments, the system accesses the three-dimensional structure of a reference amino acid sequence of the protein and fits a three-dimensional grid of voxels to the atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each atom category. Contains a voxelizer to Atoms can span multiple atomic categories specifying the atomic elements of amino acids. Each of the distance channels for each atom category may have a three-dimensional distance value for each voxel within a three-dimensional grid of voxels. The three-dimensional distance value may specify the distance from a corresponding voxel in a three-dimensional grid of voxels to an atom of a corresponding atom category within a plurality of atom categories. The system further includes an alternate allele encoder that encodes alternate allele amino acids for each voxel in the three-dimensional grid of voxels. Alternative allelic amino acids are three-dimensional representations of one-hot encoding of variant amino acids expressed by variant nucleotides. The system further includes an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels. An evolutionarily conserved sequence can be a three-dimensional representation of amino acid-specific conservation frequencies across multiple species. Amino acid-specific conservation frequencies can be selected depending on amino acid proximity to the corresponding voxel. The system further includes a tensor generator configured to generate a tensor that includes distance channels for each atomic category encoded with alternative allelic amino acids and respective evolutionarily conserved sequences.

clause set 1
1. A computer-implemented method, the method comprising:
accessing the three-dimensional structure of a reference amino acid sequence of a protein and fitting a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid;
each of the per-amino acid distance channels has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
encoding an alternative allele channel in each voxel in a three-dimensional grid of voxels, the alternative allele channel being a three-dimensional representation of a one-hot encoding of a variant amino acid expressed by a variant nucleotide; , encoding and
encoding an evolutionary conserved channel into each array of three-dimensional distance values across the per-amino acid distance channel on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
applying a three-dimensional convolution to a tensor containing per-amino acid distance channels encoded with alternative allele channels and respective evolutionary conserved channels;
determining pathogenicity of a mutant nucleotide based at least in part on the tensor;
2. 2. The computer-implemented method of clause 1, further comprising centering the three-dimensional grid of voxels on the alpha carbon atom of each residue of the reference amino acid within the reference amino acid sequence.
3. 3. The computer-implemented method of clause 2, further comprising centering the three-dimensional grid of voxels on the alpha carbon atom of a particular reference amino acid residue that corresponds to the variant amino acid.
4. Encoding the directionality of a reference amino acid and the position of a particular reference amino acid within a reference amino acid sequence by multiplying the three-dimensional distance value for the reference amino acid preceding the particular reference amino acid by a directionality parameter in a tensor. The computer-implemented method of clause 3, further comprising:
5. 5. The computer-implemented method of clause 4, wherein the distance is the nearest atom distance from the corresponding voxel center in the three-dimensional grid of voxels to the nearest atom of the corresponding reference amino acid.
6. 6. The computer-implemented method of clause 5, wherein the distance between the closest atoms is the Euclidean distance.
7. 7. The computer-implemented method of clause 6, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
8. 6. The computer-implemented method of clause 5, wherein the reference amino acid has an alpha carbon atom, and the distance is the distance of the nearest alpha carbon atom from the corresponding voxel center to the nearest alpha carbon atom of the corresponding reference amino acid.
9. 6. The computer-implemented method of clause 5, wherein the reference amino acid has a beta carbon atom, and the distance is the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding reference amino acid.
10. 6. The computer-implemented method of clause 5, wherein the reference amino acid has a backbone atom, and the distance is the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding reference amino acid.
11. 6. The computer-implemented method of clause 5, wherein the amino acid has a side chain atom and the distance is the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding reference amino acid.
12. The computer-implemented method of clause 3, further comprising encoding, in the tensor, a nearest atom channel specifying the distance from each voxel to the amino acid and the nearest atom selected regardless of the atomic element of the amino acid. .
13. 13. The computer-implemented method of clause 12, wherein the distance is a Euclidean distance.
14. 14. The computer-implemented method of clause 13, wherein the distance is normalized by dividing the Euclidean distance by the maximum distance.
15. 13. The computer-implemented method of clause 12, wherein the amino acids include non-standard amino acids.
16. 2. The computer-implemented method of clause 1, wherein the tensor further includes an absent atom channel that specifies atoms not found within a predetermined radius of the voxel center.
17. 17. The computer-implemented method of clause 16, wherein the absent atomic channels are one-hot encoded.
18. 2. The computer-implemented method of clause 1, further comprising voxel-wise encoding a reference allele channel in each voxel in the three-dimensional grid of voxels.
19. 19. The computer-implemented method of clause 18, wherein the reference allele amino acid is a three-dimensional representation of a one-hot encoding of the reference amino acid experiencing variant amino acids.
20. 2. The computer-implemented method of clause 1, wherein the amino acid-specific conservation frequency specifies the level of conservation of each amino acid across multiple species.
21.
selecting the closest atom to the corresponding voxel across the reference amino acid and atom category;
selecting a pan-amino acid conservation frequency for the residues of the reference amino acid containing the closest atoms;
21. The computer-implemented method of clause 20, further comprising: using a three-dimensional representation of pan-amino acid conservation frequencies as an evolutionary conservation channel.
22. 22. The computer-implemented method of clause 21, wherein pan-amino acid conservation frequencies are constructed for particular positions of residues as observed in multiple species.
23. 22. The computer-implemented method of clause 21, wherein the pan-amino acid conservation frequency specifies whether there is a missing conservation frequency for a particular reference amino acid.
24.
selecting the respective nearest atom to the corresponding voxel in each of the reference amino acids;
For each residue of the reference amino acid containing the nearest atom, selecting a conservation frequency for each amino acid;
22. The computer-implemented method of clause 21, further comprising: using a three-dimensional representation of conservation frequency for each amino acid as an evolutionary conservation channel.
25. 25. The computer-implemented method of clause 24, wherein a per-amino acid conservation frequency is constructed for a particular position of the residue as observed in multiple species.
26. 25. The computer-implemented method of clause 24, wherein the conserved frequency for each amino acid specifies whether there is a conserved frequency that is missing for a particular reference amino acid.
27. The computer implementation of clause 1, further comprising encoding, voxel by voxel, one or more annotation channels that are a three-dimensional representation of a one-hot encoding of residue annotations in each voxel in the three-dimensional grid of voxels. Method.
28. 28. The computer-implemented method of clause 27, wherein the annotation channel is a molecular processing annotation that includes an initiator methionine, a signal, a transit peptide, a propeptide, a chain, and a peptide.
29. Clause where the annotation channel is a region annotation including topological domain, transmembrane, intramembrane, domain, repeat, calcium binding, zinc finger, deoxyribonucleic acid (DNA) binding, nucleotide binding, region, coiled coil, motif, and compositional bias. 28. The computer implementation method according to 27.
30. 28. The computer-implemented method of clause 27, wherein the annotation channel is a site annotation that includes an active site, a metal binding, a binding site, and a site.
31. 28. The computer-implemented method of clause 27, wherein the annotation channel is an amino acid modification annotation including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks.
32. 28. The computer-implemented method of clause 27, wherein the annotation channel is a secondary structure annotation including helices, turns, and beta strands.
33. 28. The computer-implemented method of clause 27, wherein the annotation channel is an experimental information annotation that includes mutagenesis, sequence uncertainty, sequence conflicts, non-adjacent residues, and non-terminal residues.
34. Clause further comprising encoding on a voxel-by-voxel basis to each voxel in the three-dimensional grid of voxels one or more structural confidence channels that are three-dimensional representations of confidence scores specifying the quality of each residue structure. 1. The computer implementation method according to 1.
35. 35. The computer-implemented method of clause 34, wherein the structural confidence channel is a global model quality estimation (GMQE).
36. 35. The computer-implemented method of clause 34, wherein the structural confidence channel is a Qualitative Model Energy Analysis (QMEAN) score.
37. 35. The computer-implemented method of clause 34, wherein the structure confidence channel is a temperature factor that specifies the degree to which the residues meet the physical constraints of the respective protein structure.
38. 35. The computer-implemented method of clause 34, wherein the structure confidence channel is a template structure alignment that specifies the degree to which the residues of atoms closest to the voxel have aligned template structures.
39. 39. The computer-implemented method of clause 38, wherein the structure confidence channel is a template modeling score of the aligned template structure.
40. 40. The computer-implemented method of clause 39, wherein the structural confidence channels are a minimum of the template modeling scores, an average of the template modeling scores, and a maximum of the template modeling scores.
41. 2. The computer-implemented method of clause 1, further comprising rotating the atoms before the per-amino acid distance channels are generated.
42. In convolutional neural networks, use 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer The computer-implemented method of clause 1, further comprising:
43. The computer-implemented method of clause 42, wherein the 1x1x1 convolution and the 3x3x3 convolution are three-dimensional convolutions.
44. A 1x1x1 convolutional layer processes the tensor and produces an intermediate output that is a convolved representation of the tensor, and an array of 3x3x3 convolutional layers processes the intermediate output and produces a flattened representation of the tensor. A fully connected layer processes the smoothed output and generates an unnormalized output, a softmax classification layer processes the unnormalized output, and a softmax classification layer processes the unnormalized output to distinguish between pathogenic and benign variant nucleotides. 43. The computer-implemented method of clause 42, generating an exponentially normalized output that identifies a likelihood that is.
45. 45. The computer-implemented method of clause 44, wherein the sigmoid layer processes the denormalized output and generates a normalized output that identifies the likelihood that the variant nucleotide is pathogenic.
46. The computer-implemented method of clause 1, wherein the convolutional neural network is an attention-based neural network.
47. 2. The computer-implemented method of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with a reference allele channel.
48. 2. The computer-implemented method of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with an annotation channel.
49. 2. The computer-implemented method of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with a structural confidence channel.
50. A computer-implemented method, the method comprising:
By accessing the 3D structure of a protein's reference amino acid sequence and fitting a 3D grid of voxels to atoms spanning multiple atomic categories in the 3D structure on an amino acid basis, we generate distance channels for each atomic category. There it is,
Atoms span multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atom category has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionarily conserved channel into each array of three-dimensional distance values over the distance channel for each atomic category on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
applying a three-dimensional convolution to a tensor containing distance channels for each atomic category encoded with alternative allele channels and respective evolutionary conservation channels;
determining pathogenicity of a mutant nucleotide based at least in part on the tensor;
51. A computer-implemented method, the method comprising:
accessing the three-dimensional structure of a reference amino acid sequence of a protein and fitting a three-dimensional grid of voxels to the atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid;
each of the per-amino acid distance channels has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionary conserved channel into each array of three-dimensional distance values across the per-amino acid distance channel on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
an evolutionary conservation encoder in which amino acid-specific conservation frequencies are selected depending on amino acid proximity to corresponding voxels;
generating a tensor comprising per-amino acid distance channels encoded with alternative allele channels and respective evolutionary conserved channels.
52. A computer-implemented method, the method comprising:
By accessing the 3D structure of a protein's reference amino acid sequence and, on an amino acid basis, fitting a 3D grid of voxels to atoms spanning multiple atomic categories in the 3D structure, generating distance channels for each atomic category. There it is,
Atoms span multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atom category has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionarily conserved channel into each array of three-dimensional distance values over the distance channel for each atomic category on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
generating a tensor comprising distance channels for each atomic category encoded with alternative allele channels and respective evolutionary conserved channels.

clause set 2
1. When executed on one or more processors,
accessing the three-dimensional structure of a reference amino acid sequence of a protein and fitting a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid;
each of the per-amino acid distance channels has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionary conserved channel into each array of three-dimensional distance values across the per-amino acid distance channel on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
applying a three-dimensional convolution to a tensor containing per-amino acid distance channels encoded with alternative allele channels and respective evolutionary conserved channels;
one or more computer-readable media storing computer-executable instructions for configuring a computer to perform operations comprising: determining the pathogenicity of a mutant nucleotide based at least in part on the tensor;
2. 2. The computer-readable medium of clause 1, wherein the act further comprises centering the three-dimensional grid of voxels on the alpha carbon atom of each residue of the reference amino acid within the reference amino acid sequence.
3. 3. The computer-readable medium of clause 2, wherein the act further comprises centering the three-dimensional grid of voxels on the alpha carbon atom of a particular reference amino acid residue that corresponds to the variant amino acid.
4. The operation encodes the directionality of the reference amino acid and the position of the particular reference amino acid within the reference amino acid sequence by multiplying the three-dimensional distance value for the reference amino acid preceding the particular reference amino acid by the directionality parameter in the tensor. 3. The computer readable medium of clause 3, further comprising:
5. 5. The computer-readable medium of clause 4, wherein the distance is the nearest atom distance from the corresponding voxel center in the three-dimensional grid of voxels to the nearest atom of the corresponding reference amino acid.
6. 6. The computer-readable medium of clause 5, wherein the distance between the nearest atoms is the Euclidean distance.
7. 7. The computer-readable medium of clause 6, wherein the nearest atom distance is normalized by dividing the Euclidean distance by the largest nearest atom distance.
8. If the reference amino acid has an alpha carbon atom and the distance is
6. The computer-readable medium of clause 5, which is the nearest alpha carbon atom distance from the corresponding voxel center to the nearest alpha carbon atom of the corresponding reference amino acid.
9. 6. The computer-readable medium of clause 5, wherein the reference amino acid has a beta carbon atom, and the distance is the distance of the nearest beta carbon atom from the corresponding voxel center to the nearest beta carbon atom of the corresponding reference amino acid.
10. 6. The computer-readable medium of clause 5, wherein the reference amino acid has a backbone atom, and the distance is the distance of the nearest backbone atom from the corresponding voxel center to the nearest backbone atom of the corresponding reference amino acid.
11. 6. The computer-readable medium of clause 5, wherein the amino acid has a side chain atom, and the distance is the distance of the nearest side chain atom from the corresponding voxel center to the nearest side chain atom of the corresponding reference amino acid.
12. 3, wherein the operation further comprises encoding, in the tensor, a nearest atom channel specifying the distance from each voxel to the nearest atom selected without regard to the amino acid and the atomic element of the amino acid. computer readable medium.
13. 13. The computer-readable medium of clause 12, wherein the distance is a Euclidean distance.
14. 14. The computer-readable medium of clause 13, wherein the distance is normalized by dividing the Euclidean distance by the maximum distance.
15. 13. The computer-readable medium of clause 12, wherein the amino acids include non-standard amino acids.
16. 2. The computer-readable medium of clause 1, wherein the tensor further includes an absent atom channel that specifies atoms not found within a predetermined radius of the voxel center.
17. 17. The computer-readable medium of clause 16, wherein the absent atomic channel is one-hot encoded.
18. 2. The computer-readable medium of clause 1, wherein the operations further include voxel-wise encoding a reference allele channel in each voxel in the three-dimensional grid of voxels. 19. 19. The computer-readable medium of clause 18, wherein the reference allele amino acid is a three-dimensional representation of a one-hot encoding of the reference amino acid experiencing variant amino acids.
20. 2. The computer-readable medium of clause 1, wherein the amino acid-specific conservation frequency specifies the level of conservation of each amino acid across multiple species.
21. The action is
selecting the closest atom to the corresponding voxel across the reference amino acid and atom category;
selecting a pan-amino acid conservation frequency for the residues of the reference amino acid containing the closest atoms;
21. The computer-readable medium of clause 20, further comprising: using a three-dimensional representation of pan-amino acid conservation frequencies as an evolutionary conservation channel.
22. 22. The computer-readable medium of clause 21, wherein pan-amino acid conservation frequencies are constructed for particular positions of residues as observed in multiple species.
23. 22. The computer-readable medium of clause 21, wherein the pan-amino acid conservation frequency specifies whether there is a missing conservation frequency for a particular reference amino acid.
24. the operation selects each closest atom to the corresponding voxel in each of the reference amino acids;
For each residue of the reference amino acid containing the nearest atom, selecting a conservation frequency for each amino acid;
22. The computer-readable medium of clause 21, further comprising: using a three-dimensional representation of conservation frequency for each amino acid as an evolutionary conservation channel.
25. 25. The computer-readable medium of clause 24, wherein a conservation frequency for each amino acid is configured for a particular position of the residue as observed in multiple species.
26. 25. The computer-readable medium of clause 24, wherein the conserved frequency for each amino acid specifies whether there is a conserved frequency missing for a particular reference amino acid.
27. Clause 1, wherein the operation further comprises encoding, on a voxel-by-voxel basis, one or more annotation channels that are a three-dimensional representation of a one-hot encoding of residue annotations for each voxel in the three-dimensional grid of voxels. computer readable medium.
28. 28. The computer-readable medium of clause 27, wherein the annotation channel is a molecular processing annotation that includes an initiator methionine, a signal, a transit peptide, a propeptide, a chain, and a peptide.
29. Clause where the annotation channel is a region annotation including topological domain, transmembrane, intramembrane, domain, repeat, calcium binding, zinc finger, deoxyribonucleic acid (DNA) binding, nucleotide binding, region, coiled coil, motif, and compositional bias. 28. The computer readable medium according to 27.
30. 28. The computer-readable medium of clause 27, wherein the annotation channel is a site annotation that includes an active site, a metal binding, a binding site, and a site.
31. 28. The computer-readable medium of clause 27, wherein the annotation channel is an amino acid modification annotation including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks.
32. 28. The computer-readable medium of clause 27, wherein the annotation channel is a secondary structure annotation including helices, turns, and beta strands.
33. 28. The computer-readable medium of clause 27, wherein the annotation channel is an experimental information annotation that includes mutagenesis, sequence uncertainty, sequence conflicts, non-adjacent residues, and non-terminal residues.
34. The operation encodes, for each voxel, one or more structural confidence channels, which are three-dimensional representations of confidence scores that specify the quality of the respective residue structures, for each voxel in the three-dimensional grid of voxels. The computer-readable medium of clause 1, further comprising:
35. 35. The computer-readable medium of clause 34, wherein the structural confidence channel is global model quality estimation (GMQE).
36. 35. The computer-readable medium of clause 34, wherein the structural confidence channel is a Qualitative Model Energy Analysis (QMEAN) score.
37. 35. The computer-readable medium of clause 34, wherein the structure confidence channel is a temperature factor that specifies the degree to which the residues meet the physical constraints of their respective protein structure.
38. 35. The computer-readable medium of clause 34, wherein the structure confidence channel is a template structure alignment that specifies the degree to which the residues of atoms closest to the voxel have an aligned template structure.
39. 39. The computer-readable medium of clause 38, wherein the structure confidence channel is a template modeling score for aligned template structures.
40. 40. The computer-readable medium of clause 39, wherein the structural confidence channels are a minimum of the template modeling scores, an average of the template modeling scores, and a maximum of the template modeling scores.
41. 2. The computer-readable medium of clause 1, wherein the operations further include rotating the atoms before the per-amino acid distance channels are generated.
42. The operations are 1x1x1 convolution, 3x3x3 convolution, normalized linear unit activation layer, batch normalization layer, fully connected layer, dropout regularization layer, and softmax classification layer in convolutional neural networks. The computer-readable medium of clause 1, further comprising using the computer-readable medium of clause 1.
43. The computer-readable medium of clause 42, wherein the 1x1x1 convolution and the 3x3x3 convolution are three-dimensional convolutions.
44. A 1x1x1 convolutional layer processes the tensor and produces an intermediate output that is a convolved representation of the tensor, and an array of 3x3x3 convolutional layers processes the intermediate output and flattens it. a fully connected layer processes the flattened output and generates an unnormalized output, a softmax classification layer processes the unnormalized output, and determines whether the variant nucleotides are pathogenic or benign. 43. The computer-readable medium of clause 42, wherein the computer-readable medium generates an exponentially normalized output that identifies a likelihood.
45. 45. The computer-readable medium of clause 44, wherein the sigmoid layer processes the denormalized output and produces a normalized output that identifies a likelihood that the variant nucleotide is pathogenic.
46. The computer-readable medium of clause 1, wherein the convolutional neural network is an attention-based neural network.
47. 2. The computer-readable medium of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with a reference allele channel.
48. The computer-readable medium of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with an annotation channel.
49. 2. The computer-readable medium of clause 1, wherein the tensor includes a per-amino acid distance channel further encoded with a structural confidence channel.
50. When executed on one or more processors:
By accessing the 3D structure of a protein's reference amino acid sequence and fitting a 3D grid of voxels to atoms spanning multiple atomic categories in the 3D structure on an amino acid basis, we generate distance channels for each atomic category. There it is,
Atoms span multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atom category has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionarily conserved channel into each array of three-dimensional distance values over the distance channel for each atomic category on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
applying a three-dimensional convolution to a tensor containing distance channels for each atomic category encoded with alternative allele channels and respective evolutionary conservation channels;
one or more computer-readable media storing computer-executable instructions for configuring a computer to perform operations comprising: determining the pathogenicity of a mutant nucleotide based at least in part on the tensor;
51. When executed on one or more processors:
accessing the three-dimensional structure of a reference amino acid sequence of a protein and fitting a three-dimensional grid of voxels to the atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each amino acid;
each of the per-amino acid distance channels has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
On an amino acid position basis, each three-dimensional distance value in each of the per-amino acid distance channels is encoded with an alternative allele channel that is a one-hot encoded three-dimensional representation of the variant amino acids expressed by the variant nucleotides. And,
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionary conserved channel into each array of three-dimensional distance values across the per-amino acid distance channel on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
storing computer-executable instructions configuring the computer to perform operations comprising: generating a tensor comprising per-amino acid distance channels encoded with alternative allele channels and respective evolutionary conserved channels; one or more computer readable media.
52. When executed on one or more processors:
By accessing the 3D structure of a protein's reference amino acid sequence and, on an amino acid basis, fitting a 3D grid of voxels to atoms spanning multiple atomic categories in the 3D structure, generating distance channels for each atomic category. There it is,
Atoms span multiple atomic categories,
an atom category of the plurality of atom categories specifies an atomic element of an amino acid,
each of the distance channels for each atom category has a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
generating a three-dimensional distance value specifying a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
encoding an alternative allele channel into each voxel in a three-dimensional grid of voxels;
encoding, wherein the alternative allele channel is a three-dimensional representation of a one-hot encoding of variant amino acids expressed by variant nucleotides;
encoding an evolutionarily conserved channel into each array of three-dimensional distance values over the distance channel for each atomic category on a voxel location basis,
an evolutionary conservation channel is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
encoding, wherein amino acid-specific conservation frequencies are selected according to amino acid proximity to corresponding voxels;
generating a tensor comprising distance channels for each atomic category encoded with alternative allele channels and respective evolutionary conserved channels; One or more computer readable media for storage.

Other implementations of the methods described in this section may include a non-transitory computer-readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another embodiment of the methods described in this section includes a memory and one or more processors operable to execute instructions stored in the memory to perform any of the methods described above. A system that includes:

Particular embodiment 3
Clause Set 1
1. A computer-implemented method for efficiently determining an element of an array having element coordinates that is closest to a cell having a cell index and cell coordinates per dimension, the cell index and cell coordinates being uniformly spaced within a grid, comprising:
generating an element-to-cell mapping that maps a subset of the cells to each of the elements;
the subset of cells mapped to a particular element in the array includes the nearest cell in the grid and one or more neighboring cells in the grid;
The closest cell is selected based on matching the element coordinates of the particular element to the cell coordinates;
generating a neighborhood cell, the neighborhood cells being selected based on being contiguous adjacent to the nearest cell and within a distance proximity range from the particular element;
generating a cell-to-element mapping that maps a subset of the elements to each of the cells;
generating a subset of elements that map to particular voxels in the grid, the subset including elements in the array that map to particular cells according to an element-to-cell mapping;
determining, for each of the cells, the closest element in the array using a cell-to-element mapping;
determining, where a closest element to the particular cell is determined based on a distance between the particular cell and an element in the subset of elements.
2. The computer-implemented method of clause 1, wherein matching element coordinates of a particular element to cell coordinates further comprises truncating a fractional portion of the element coordinates to generate truncated element coordinates.
3. Matching the element coordinates of a specific element to the cell coordinates
For a first dimension, matching a first truncated element coordinate in the truncated element coordinates to a first cell coordinate of a first cell in the grid and selecting a first dimension index of the first cell;
For a second dimension, matching a second truncated element coordinate in the truncated element coordinates to a second cell coordinate of a second cell in the grid and selecting a second dimension index of the second cell;
For a third dimension, matching a third truncated element coordinate in the truncated element coordinates to a third cell coordinate of a third cell in the grid and selecting a third dimension index of the third cell;
using the selected first, second, and third dimension indices to generate a cumulative sum based on position-wise weighting the selected first, second, and third dimension indices by a power of the radix;
3. The computer-implemented method of claim 2, further comprising: using the cumulative sum as a cell index for selection of the closest cell.
4. The computer-implemented method of claim 1, wherein the distance is calculated between a cell coordinate of a particular cell and an element coordinate of an element in the subset of elements.
5. The computer-implemented method of claim 1, wherein the sequence is a protein sequence of amino acids.
6. The computer-implemented method of claim 5, wherein the elements are atoms of an amino acid.
7. The steps of generating a mapping from elements to cells, generating a mapping from cells to elements, and determining, for each of the cells, the closest element using the mapping from cells to elements has a runtime complexity of O(a ^* f+v);
a is the number of atoms,
f is the number of amino acids,
v is the number of cells,
7. The computer-implemented method of claim 6, wherein ^* is a multiplication operation.
8. The computer-implemented method of claim 7, wherein the atom comprises an alpha carbon atom.
9. The computer-implemented method of claim 7, wherein the atom comprises a beta carbon atom.
10. The computer-implemented method of claim 7, wherein the atoms include non-carbon atoms.
11. The computer-implemented method of claim 1, wherein the cells are three-dimensional voxels.
12. The computer-implemented method of claim 11, wherein the cell coordinates are three-dimensional coordinates.
13. The computer-implemented method of claim 12, wherein the element coordinates are three-dimensional coordinates.
14. The computer-implemented method of claim 1, wherein the neighboring cells are selected based on being within an index contiguous range from a nearest cell.
15. The computer-implemented method of claim 1, wherein the neighboring cell is selected based on being within a neighborhood of a cell in a grid that includes the nearest cell.
16. The computer-implemented method of clause 1, wherein the array includes M elements and the subset of elements includes N elements, where M>>N.
17. A computer-implemented method for efficiently determining an atom having three-dimensional (3D) atomic coordinates in a protein that is closest to a voxel having 3D voxel coordinates in a grid, comprising:
generating an atom-to-voxel mapping that maps a containing voxel selected based on matching 3D atomic coordinates of a particular atom of the protein to 3D voxel coordinates in the grid to each of the atoms;
1. A computer-implemented method comprising: generating a voxel-to-atom mapping that maps a subset of atoms to each of the voxels, wherein the subset of atoms that are mapped to a particular voxel in the grid includes atoms in the protein that are mapped to the particular voxel by the atom-to-voxel mapping; and determining, for each of the voxels, a closest atom in the protein using the voxel-to-atom mapping.
18. The computer-implemented method of claim 17, wherein the steps of clause 17 have a runtime complexity of 0 (number of atoms).

Other implementations of the methods described in this section may include a non-transitory computer-readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another embodiment of the methods described in this section includes a memory and one or more processors operable to execute instructions stored in the memory to perform any of the methods described above. A system that includes:

Although the invention has been disclosed with reference to the above-described preferred embodiments and examples, it is to be understood that these examples are intended in an illustrative rather than a restrictive sense. It is. Modifications and combinations will readily occur to those skilled in the art and are considered to be within the spirit of the invention and the scope of the appended claims.

104 Sequence Accessor 114 3D Structure Generator 124 Coordinate Classifier 134 Voxel Grid Generator 144 Voxel Grid Centerer 154 Distance Channel Generator 164 One Hot Encoder 174 Concatenator 184 Runtime Logic 1204 Similar Sequence Finder 1214 Aligner 1224 Pan-Amino Acid Conservation Frequency Calculator 1234 Nearest atom finder 1244 Amino acid selector 1254 Voxelizer 1924 Preservation frequency calculator for each amino acid 1934 Nearest atom finder 1944 Amino acid selector 1954 Voxelizer 2108 Pathogenicity classifier 3600 Computer system 3610 Memory subsystem 3622 Memory subsystem 3632 RAM
3634 ROM
3636 File Storage Subsystem 3638 User Interface Input Device 3655 Bus Subsystem 3672 CPU
3674 Network Interface Subsystem 3676 User Interface Output Device 3678 Processor (GPU, FPGA, CGRA)

Claims (52)

A system,
A voxelizer that accesses a three-dimensional structure of a reference amino acid sequence of a protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate a distance channel for each amino acid, comprising:
each of the per-amino acid distance channels having a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
a voxelizer, wherein the three-dimensional distance value specifies the distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
an alternative allele encoder encoding alternative allele amino acids in each voxel in the three-dimensional grid of voxels, the alternative allele encoder comprising:
an alternative allele encoder, wherein the alternative allele amino acids are one-hot encoded three-dimensional representations of variant amino acids expressed by variant nucleotides;
an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels, the encoder comprising:
the evolutionary conserved sequence is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
an evolutionary conservation encoder, wherein the amino acid-specific conservation frequency is selected according to amino acid proximity to the corresponding voxel;
A convolutional neural network,
applying a three-dimensional convolution to a tensor containing the per-amino acid distance channels encoded by the alternative allele amino acids and their respective evolutionarily conserved sequences;
a convolutional neural network configured to determine pathogenicity of the mutant nucleotide based at least in part on the tensor. The system of claim 1, wherein the voxelizer centers the three-dimensional grid of voxels on the alpha carbon atom of each residue of a reference amino acid in the reference amino acid sequence. 3. The system of claim 1 or 2, wherein the voxelizer centers the three-dimensional grid of voxels on the alpha carbon atom of a particular reference amino acid residue located in the variant amino acid. In the tensor, the directionality of the reference amino acid and the position of the particular reference amino acid within the reference amino acid sequence are determined by multiplying the three-dimensional distance value for the reference amino acid preceding the particular reference amino acid by a directionality parameter. 4. A system according to any one of claims 1 to 3, further configured to encode. 5. The system of any one of claims 1 to 4, wherein the distance is the nearest atomic distance from the corresponding voxel center in the three-dimensional grid of voxels to the nearest atom of the corresponding reference amino acid. 6. The system of claim 5, wherein the closest atomic distance is a Euclidean distance. 7. The system of claim 5 or 6, wherein the nearest atomic distance is normalized by dividing the Euclidean distance by the largest nearest atomic distance. 8. The reference amino acid of claim 1 to 7, wherein the reference amino acid has an alpha carbon atom, and the distance is the nearest alpha carbon atom distance from the corresponding voxel center to the nearest alpha carbon atom of the corresponding reference amino acid. A system according to any one of the clauses. 8. The reference amino acid of claim 1 to 7, wherein the reference amino acid has a beta carbon atom, and the distance is the nearest beta carbon atom distance from the corresponding voxel center to the nearest beta carbon atom of the corresponding reference amino acid. A system according to any one of the clauses. Any one of claims 1 to 7, wherein the reference amino acid has a backbone atom, and the distance is the nearest backbone atom distance from the corresponding voxel center to the nearest backbone atom of the corresponding reference amino acid. The system described in Section. 8. The reference amino acid of claim 1 to 7, wherein the reference amino acid has a side chain atom, and the distance is the nearest side chain atom distance from the corresponding voxel center to the nearest side chain atom of the corresponding reference amino acid. A system according to any one of the clauses. in the tensor, further configured to encode a nearest atom channel specifying a distance from each voxel to the nearest atom, the nearest atom being selected regardless of the amino acid and the atomic element of the amino acid; 12. A system according to any one of claims 1 to 11. 13. A system according to any preceding claim, wherein the distance is a Euclidean distance. 14. A system according to any preceding claim, wherein the distance is normalized by dividing the Euclidean distance by a maximum distance. 15. The system according to any one of claims 1 to 14, wherein the amino acids include non-standard amino acids. The system of any one of claims 1 to 15, wherein the tensor further includes absent atom channels that specify atoms that are not found within a predetermined radius of a voxel center. 17. The system of claim 16, wherein the absent atomic channel is one-hot encoded. 18. The system of any one of claims 1 to 17, further comprising a reference allele encoder for encoding reference allele amino acids on a voxel-by-voxel basis in each voxel in the three-dimensional grid of voxels. 19. The system of claim 18, wherein the reference allele amino acid is a one-hot encoded three-dimensional representation of the reference amino acid that experiences the variant amino acid. 20. The system of any one of claims 1-19, wherein the amino acid-specific conservation frequency specifies the level of conservation of each amino acid across the plurality of species. The evolutionarily conserved encoder comprises:
selecting the closest atom to the corresponding voxel across the reference amino acid and the atom category;
Select a pan-amino acid conservation frequency for the residue of the reference amino acid that contains the closest atom;
The system according to any one of claims 1 to 20, wherein a three-dimensional representation of the general amino acid conservation frequency is used as the evolutionarily conserved sequence. 22. The system of claim 21, wherein the pan-amino acid conservation frequency is constructed for a particular position of the residue as observed in the plurality of species. 23. The system of claim 21 or 22, wherein the pan-amino acid conservation frequency identifies whether a missing conservation frequency exists for a particular reference amino acid. The evolutionary conserving encoder is
selecting the respective nearest atom to the corresponding voxel in each of the reference amino acids;
For each residue of the reference amino acid containing the nearest atom, select a conservation frequency for each amino acid;
24. The system according to any one of claims 1 to 23, wherein a three-dimensional representation of the conservation frequency for each amino acid is used as the evolutionarily conserved sequence. 25. The system of claim 24, wherein the per-amino acid conservation frequency is configured for a particular position of a residue as observed in the plurality of species. The system of claim 24 or 25, wherein the conservation frequency for each amino acid identifies whether a missing conservation frequency exists for a particular reference amino acid. further comprising an annotation encoder that encodes one or more annotation channels on a voxel-by-voxel basis for each voxel in the three-dimensional grid of voxels;
27. The system of any one of claims 1 to 26, wherein the annotation channel is a three-dimensional representation of a one-hot encoding of residue annotations. 28. The system of claim 27, wherein the annotation channel is a molecular processing annotation that includes initiator methionine, signal, transit peptide, propeptide, chain, and peptide. the annotation channel is a region annotation including topological domain, transmembrane, intramembrane, domain, repeat, calcium binding, zinc finger, deoxyribonucleic acid (DNA) binding, nucleotide binding, region, coiled coil, motif, and compositional bias; A system according to claim 27 or 28. 30. The system of any one of claims 27 to 29, wherein the annotation channel is a site annotation comprising an active site, a metal binding, a binding site, and a site. 31. The system of any one of claims 27-30, wherein the annotation channel is an amino acid modification annotation including non-standard residues, modified residues, lipidations, glycosylation, disulfide bonds, and crosslinks. The system of any one of claims 27 to 31, wherein the annotation channel is a secondary structure annotation including helices, turns, and beta strands. The system of any one of claims 27 to 32, wherein the annotation channel is an experimental information annotation including mutagenesis, sequence uncertainty, sequence conflicts, non-adjacent residues, and non-terminal residues. a structure confidence encoder for encoding one or more structure confidence channels for each voxel in the three-dimensional grid of voxels on a voxel-by-voxel basis;
34. The system of claim 1, wherein the structure confidence channel is a three-dimensional representation of confidence scores that specify the quality of each residue structure. 35. The system of claim 34, wherein the structural confidence channel is global model quality estimation (GMQE). 36. The system of claim 34 or 35, wherein the structural confidence channel is a Qualitative Model Energy Analysis (QMEAN) score. 37. The system of any one of claims 34-36, wherein the structure confidence channel is a temperature factor that specifies the degree to which the residues meet the physical constraints of their respective protein structure. 38. The system of any one of claims 34-37, wherein the structure confidence channel is a template structure alignment that specifies the degree to which the closest atomic residues to the voxel have an aligned template structure. 39. The system of any one of claims 34-38, wherein the structure confidence channel is a template modeling score of the aligned template structure. The system of claim 39, wherein the structural confidence channel is a minimum one of the template modeling scores, an average of the template modeling scores, and a maximum one of the template modeling scores. 41. The system of any one of claims 1-40, further comprising an atomic rotation engine that rotates the atoms before the per-amino acid distance channels are generated. The convolutional neural network uses a 1×1×1 convolution, a 3×3×3 convolution, a normalized linear unit activation layer, a batch normalization layer, a fully connected layer, a dropout regularization layer, and a softmax classification layer. 42. The system according to any one of claims 1 to 41, wherein: 43. The system of claim 42, wherein the 1x1x1 convolution and the 3x3x3 convolution are the three-dimensional convolutions. the 1×1×1 convolutional layer processes the tensor and produces an intermediate output that is a convolved representation of the tensor;
The array of 3x3x3 convolutional layers produces a flattened output, the fully connected layer processes the flattened output and produces a denormalized output, and the softmax classification 44. The system of claim 42 or 43, wherein a layer processes the denormalized output and produces an exponentially normalized output that identifies the likelihood that the variant nucleotide is pathogenic and benign. . 45. The system of claim 44, wherein a sigmoid layer processes the denormalized output and generates a normalized output that identifies the likelihood that the variant nucleotide is pathogenic. The system of any one of claims 1 to 45, wherein the convolutional neural network is an attention-based neural network. 47. The system of any one of claims 1 to 46, wherein the tensor comprises a distance channel for each of the amino acids further encoded with the reference allele amino acids. 48. The system of any one of claims 27-47, wherein the tensor comprises a distance channel for each of the amino acids further encoded with the annotation channel. 49. The system of any one of claims 34-48, wherein the tensor comprises the per-amino acid distance channel further encoded with the structure confidence channel. A system,
A voxelizer that accesses the three-dimensional structure of a reference amino acid sequence of a protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate distance channels for each atom category,
the atom spans multiple atomic categories,
an atomic category of the plurality of atomic categories specifies an atomic element of the amino acid;
each of the distance channels for each atomic category having a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
a voxelizer, wherein the three-dimensional distance value specifies a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
an alternative allele encoder encoding alternative allele amino acids in each voxel in the three-dimensional grid of voxels, the alternative allele encoder comprising:
an alternative allele encoder, wherein the alternative allele amino acids are one-hot encoded three-dimensional representations of variant amino acids expressed by variant nucleotides;
an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels, the encoder comprising:
the evolutionary conserved sequence is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species,
an evolutionary conservation encoder, wherein the amino acid-specific conservation frequency is selected according to amino acid proximity to the corresponding voxel;
A convolutional neural network,
applying a three-dimensional convolution to a tensor containing distance channels for each of the atomic categories encoded with the alternative allele amino acids and their respective evolutionarily conserved sequences;
a convolutional neural network configured to determine pathogenicity of the mutant nucleotide based at least in part on the tensor. A system,
A voxelizer that accesses the three-dimensional structure of a reference amino acid sequence of a protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate a distance channel for each amino acid,
each of the per-amino acid distance channels having a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
a voxelizer, wherein the three-dimensional distance value specifies the distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding reference amino acid in the reference amino acid sequence;
an alternative allele encoder encoding alternative allele amino acids in each voxel in the three-dimensional grid of voxels, the alternative allele encoder comprising:
an alternative allele encoder, wherein the alternative allele amino acids are one-hot encoded three-dimensional representations of variant amino acids expressed by variant nucleotides;
an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels, the encoder comprising:
the evolutionary conserved sequence is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species;
an evolutionary conservation encoder, wherein the amino acid-specific conservation frequency is selected according to amino acid proximity to the corresponding voxel;
a tensor generator configured to generate a tensor comprising a distance channel for each of the amino acids encoded with the alternative allelic amino acids and their respective evolutionarily conserved sequences. A system,
A voxelizer that accesses the three-dimensional structure of a reference amino acid sequence of a protein and fits a three-dimensional grid of voxels to atoms in the three-dimensional structure on an amino acid basis to generate a distance channel for each atom category, the voxelizer comprising: Atoms span multiple atomic categories,
an atomic category of the plurality of atomic categories specifies an atomic element of the amino acid;
each of the distance channels for each atomic category having a three-dimensional distance value for each voxel in the three-dimensional grid of voxels;
a voxelizer, wherein the three-dimensional distance value specifies a distance from a corresponding voxel in the three-dimensional grid of voxels to an atom of a corresponding atom category in the plurality of atom categories;
an alternative allele encoder encoding alternative allele amino acids in each voxel in the three-dimensional grid of voxels, the alternative allele encoder comprising:
an alternative allele encoder, wherein the alternative allele amino acids are one-hot encoded three-dimensional representations of variant amino acids expressed by variant nucleotides;
an evolutionarily conserved encoder that encodes an evolutionarily conserved sequence in each voxel in the three-dimensional grid of voxels, the encoder comprising:
the evolutionary conserved sequence is a three-dimensional representation of amino acid-specific conservation frequencies across multiple species,
an evolutionary conservation encoder, wherein the amino acid-specific conservation frequency is selected according to amino acid proximity to the corresponding voxel;
a tensor generator configured to generate a tensor comprising a distance channel for each of the atomic categories encoded with the alternative allelic amino acids and respective evolutionarily conserved sequences.