WO2020081122A1 - Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond - Google Patents

Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond Download PDF

Info

Publication number
WO2020081122A1
WO2020081122A1 PCT/US2019/031621 US2019031621W WO2020081122A1 WO 2020081122 A1 WO2020081122 A1 WO 2020081122A1 US 2019031621 W US2019031621 W US 2019031621W WO 2020081122 A1 WO2020081122 A1 WO 2020081122A1
Authority
WO
WIPO (PCT)
Prior art keywords
training
supplemental
missense
sequence pairs
training example
Prior art date
Application number
PCT/US2019/031621
Other languages
English (en)
Inventor
Hong Gao
Kai-How FARH
Samskruthi REDDY PADIGEPATI
Original Assignee
Illumina, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from PCT/US2018/055840 external-priority patent/WO2019079166A1/fr
Priority claimed from US16/407,149 external-priority patent/US10540591B2/en
Priority to CN201980003263.9A priority Critical patent/CN111328419B/zh
Priority to EP19729404.4A priority patent/EP3659143B1/fr
Priority to JP2019567603A priority patent/JP6888123B2/ja
Priority to EP23209885.5A priority patent/EP4296899A3/fr
Priority to KR1020197038080A priority patent/KR102165734B1/ko
Priority to NZ759665A priority patent/NZ759665A/en
Application filed by Illumina, Inc. filed Critical Illumina, Inc.
Priority to SG11201911777QA priority patent/SG11201911777QA/en
Priority to CA3064223A priority patent/CA3064223A1/fr
Priority to AU2019272062A priority patent/AU2019272062B2/en
Priority to IL271091A priority patent/IL271091B/en
Publication of WO2020081122A1 publication Critical patent/WO2020081122A1/fr
Priority to IL282689A priority patent/IL282689A/en
Priority to AU2021269351A priority patent/AU2021269351B2/en

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/20—Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/30—Detection of binding sites or motifs
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B30/00—ICT specially adapted for sequence analysis involving nucleotides or amino acids
    • G16B30/10—Sequence alignment; Homology search
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
    • G16B40/20—Supervised data analysis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B50/00—ICT programming tools or database systems specially adapted for bioinformatics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections

Definitions

  • Document 3 F. Yu and V. Koltun,“MULTI-SCALE CONTEXT AGGREGATION BY DILATED CONVOLUTIONS,” arXiv: 1511.07122, 2016;
  • Document 4 K. He, X. Zhang, S. Ren, and J. Sun,“DEEP RESIDUAL LEARNING FOR IMAGE RECOGNITION,” arXiv: 1512.03385, 2015;
  • Document 1 describes deep convolutional neural network architectures that use groups of residual blocks with convolution filters having same convolution window size, batch normalization layers, rectified linear unit (abbreviated ReLU) layers, dimensionality altering layers, atrous convolution layers with exponentially growing atrous convolution rates, skip connections, and a softmax classification layer to accept an input sequence and produce an output sequence that scores entries in the input sequence.
  • the technology disclosed uses neural network components and parameters described in Document 1.
  • the technology disclosed modifies the parameters of the neural network components described in Document 1. For instance, unlike in Document 1, the atrous convolution rate in the technology disclosed progresses non-exponentially from a lower residual block group to a higher residual block group.
  • the convolution window size in the technology disclosed varies between groups of residual blocks.
  • Document 2 describes details of the deep convolutional neural network architectures described in
  • Atrous convolutions are also referred to as“dilated convolutions”. Atrous/dilated convolutions allow for large receptive fields with few trainable parameters.
  • An atrous/dilated convolution is a convolution where the kernel is applied over an area larger than its length by skipping input values with a certain step, also called atrous convolution rate or dilation factor. Atrous/dilated convolutions add spacing between the elements of a convolution filter/kemel so that neighboring input entries (e.g., nucleotides, amino acids) at larger intervals are considered when a convolution operation is performed. This enables incorporation of long-range contextual dependencies in the input.
  • the atrous convolutions conserve partial convolution calculations for reuse as adjacent nucleotides are processed.
  • Document 4 describes residual blocks and residual connections used by the technology disclosed.
  • Document 5 describes skip connections used by the technology disclosed. As used herein, skip connections are also referred to as“highway networks”.
  • Document 6 describes densely connected convolutional network architectures used by the technology disclosed.
  • Document 7 describes dimensionality altering convolution layers and modules-based processing pipelines used by the technology disclosed.
  • One example of a dimensionality altering convolution is a 1 x 1 convolution.
  • Document 8 describes batch normalization layers used by the technology disclosed.
  • Document 9 also describes atrous/dilated convolutions used by the technology disclosed.
  • Document 10 describes various architectures of deep neural networks that can be used by the technology disclosed, including convolutional neural networks, deep convolutional neural networks, and deep convolutional neural networks with atrous/dilated convolutions.
  • Document 11 describes details of a convolutional neural network that can be used by the technology disclosed, including algorithms for training a convolutional neural network with subsampling layers (e.g., pooling) and fully -connected layers.
  • subsampling layers e.g., pooling
  • Document 12 describes details of various convolution operations that can be used by the technology disclosed.
  • Document 13 describes various architectures of convolutional neural networks that can be used by the technology disclosed.
  • the technology disclosed relates to artificial intelligence type computers and digital data processing systems and corresponding data processing methods and products for emulation of intelligence (i.e., knowledge based systems, reasoning systems, and knowledge acquisition systems); and including systems for reasoning with uncertainty (e.g., fuzzy logic systems), adaptive systems, machine learning systems, and artificial neural networks.
  • intelligence i.e., knowledge based systems, reasoning systems, and knowledge acquisition systems
  • systems for reasoning with uncertainty e.g., fuzzy logic systems
  • adaptive systems e.g., machine learning systems
  • machine learning systems e.g., machine learning systems
  • artificial neural networks e.g., a neural network that uses neural networks to avoid overfitting.
  • input variables are used to predict an output variable.
  • the output variable is often called the response or dependent variable and is denoted by the variable Y,.
  • the relationship between Y and the corresponding X can be written in a general form:
  • the error term is independent of X and has a mean value of zero.
  • f is the estimate of G , which is often considered a black box, meaning that only the relation between the input and output of f is known, but the question why it works remains unanswered.
  • the function f is found using learning.
  • Supervised learning and unsupervised learning are two ways used in machine learning for this task.
  • labeled data is used for training.
  • labels By showing the inputs and the corresponding outputs (Mabels), the function f is optimized such that it approximates the output.
  • unsupervised learning the goal is to find a hidden structure from unlabeled data. The algorithm has no measure of accuracy on the input data, which distinguishes it from supervised learning.
  • a neural network is a system of interconnected artificial neurons (e.g., ai, a 2 , a 3 ) that exchange messages between each other.
  • the illustrated neural network has three inputs, two neurons in the hidden layer and two neurons in the output layer.
  • the hidden layer has an activation function f ( ⁇ ) and the output layer has an activation function g( ⁇ ) .
  • the connections have numeric weights (e.g., wn, w 2i , Wi 2 , w 3i , w 22 , w 32 , vn, v 22 ) that are tuned during the training process, so that a properly trained network responds correctly when fed an image to recognize.
  • the input layer processes the raw input
  • the hidden layer processes the output from the input layer based on the weights of the connections between the input layer and the hidden layer.
  • the output layer takes the output from the hidden layer and processes it based on the weights of the connections between the hidden layer and the output layer.
  • the network includes multiple layers of feature-detecting neurons. Each layer has many neurons that respond to different combinations of inputs from the previous layers. These layers are constructed so that the first layer detects a set of primitive patterns in the input image data, the second layer detects patterns of patterns and the third layer detects patterns of those patterns.
  • a neural network model is trained using training samples before using it used to predict outputs for production samples.
  • the quality of predictions of the trained model is assessed by using a test set of training samples that is not given as input during training. If the model correctly predicts the outputs for the test samples then it can be used in inference with high confidence. However, if the model does not correctly predict the output for test samples then we can say that the model is overfitted on the training data and it has not been generalized on the unseen test data.
  • FIG. 1 illustrates an architectural level schematic of a system in which supplemental training examples are used to reduce overfitting during training of a variant pathogenicity prediction model.
  • FIG. 2 shows an example architecture of a deep residual network for pathogenicity prediction, referred to herein as“Primate AI”.
  • FIG. 3 depicts a schematic illustration of PrimateAI, the deep learning network architecture for pathogenicity classification.
  • FIG. 4 depicts one implementation of workings of a convolutional neural network.
  • FIG. 5 depicts a block diagram of training a convolutional neural network in accordance with one implementation of the technology disclosed.
  • FIG. 6 presents an example missense variant and corresponding supplemental benign training example.
  • FIG. 7 illustrates disclosed pre-training of the pathogenicity prediction model using supplementary datasets.
  • FIG. 8 illustrates training of the pre-trained pathogenicity prediction model after the pre-training epochs.
  • FIG. 9 illustrates application of the trained pathogenicity prediction model to evaluate unlabeled variants.
  • FIG. 10 presents position frequency matrix starting point for an example amino acid sequence with pathogenic missense variant and corresponding supplemental benign training example.
  • FIG. 11 presents position frequency matrix starting point for an example amino acid sequence with benign missense variant and corresponding supplemental benign training example.
  • FIG. 12 illustrates construction of position frequency matrices for primate, mammal, and vertebrate amino acid sequences.
  • FIG. 13 presents example one hot encoding of a human reference amino acid sequence and a human alternative amino acid sequence.
  • FIG. 14 presents examples of inputs to the variant pathogenicity prediction model.
  • FIG. 15 is a simplified block diagram of a computer system that can be used to implement the technology disclosed.
  • Supplemental training examples include the same position frequency matrices (“PFMs”) as missense training examples, which may be unlabeled (and presumed pathogenic), labeled pathogenic, or labeled benign.
  • PFMs position frequency matrices
  • the intuitive impact of these supplemental benign training examples is to force the backward propagation training to distinguish between the benign and pathogenic on a basis other than the position frequency matrix.
  • the supplemental benign training example is constructed to contrast against a pathogenic or unlabeled example in the training set.
  • the supplemental benign training example could also reinforce a benign missense example.
  • the pathogenic missense can be a curated pathogenic missense or it can be a combinatorially generated example in a training set.
  • the chosen benign variant can be a synonymous variant, expressing the same amino acid from two different codons, two different trinucleotide sequences that code for the same amino acid.
  • a synonymous benign variant is used, it is not randomly constructed; instead, it is selected from synonymous variants observed in a sequenced population.
  • the synonymous variant is likely to be a human variant, as more sequence data is available for humans than for other primates, mammals or vertebrates.
  • Supplemental benign training examples have the same amino acid sequence in both reference and alternate amino acid sequences.
  • the chosen benign variant can simply be at the same location as the training example against which it contrasts. This potentially can be as effective in counteracting overfitting as use of synonymous benign variants.
  • Use of the supplemental benign training examples can be discontinued after initial training epochs or can continue throughout training, as the examples accurately reflect nature.
  • a convolutional neural network is a special type of neural network.
  • the fundamental difference between a densely connected layer and a convolution layer is this. Dense layers learn global patterns in their input feature space, whereas convolution layers learn local patters: in the case of images, patterns found in small 2D windows of the inputs. This key characteristic gives convolutional neural networks two interesting properties: (1) the patterns they learn are translation invariant and (2) they can learn spatial hierarchies of patterns.
  • a convolution layer can recognize it anywhere, for example, in the upper-left comer. A densely connected network would have to leam the pattern anew if it appeared at a new location. This makes convolutional neural networks data efficient because they need fewer training samples to leam representations that have generalization power.
  • a first convolution layer can leam small local patterns such as edges
  • a second convolution layer will leam larger patterns made of the features of the first layers, and so on. This allows convolutional neural networks to efficiently leam increasingly complex and abstract visual concepts.
  • a convolutional neural network leams highly non-linear mappings by interconnecting layers of artificial neurons arranged in many different layers with activation functions that make the layers dependent. It includes one or more convolutional layers, interspersed with one or more sub-sampling layers and non-linear layers, which are typically followed by one or more fully connected layers. Each element of the convolutional neural network receives inputs from a set of features in the previous layer. The convolutional neural network leams concurrently because the neurons in the same feature map have identical weights. These local shared weights reduce the complexity of the network such that when multi-dimensional input data enters the network, the convolutional neural network avoids the complexity of data reconstruction in feature extraction and regression or classification process.
  • Convolutions operate over 3D tensors, called feature maps, with two spatial axes (height and width) as well as a depth axis (also called the channels axis).
  • depth axis also called the channels axis.
  • the dimension of the depth axis is 3, because the image has three color channels: red, green, and blue.
  • the depth is 1 (levels of gray).
  • the convolution operation extracts patches from its input feature map and applies the same transformation to all of these patches, producing an output feature map.
  • This output feature map is still a 3D tensor: it has a width and a height.
  • Its depth can be arbitrary, because the output depth is a parameter of the layer, and the different channels in that depth axis no longer stand for specific colors as in RGB input; rather, they stand for filters. Filters encode specific aspects of the input data: at a height level, a single filter could encode the concept“presence of a face in the input,” for instance.
  • the first convolution layer takes a feature map of size (28, 28, 1) and outputs a feature map of size (26, 26, 32): it computes 32 filters over its input.
  • Each of these 32 output channels contains a 26 x 26 grid of values, which is a response map of the filter over the input, indicating the response of that filter pattern at different locations in the input. That is what the term feature map means: every dimension in the depth axis is a feature (or filter), and the 2D tensor output [:, :, n] is the 2D spatial map of the response of this filter over the input.
  • Convolutions are defined by two key parameters: (1) size of the patches extracted from the inputs - these are typically l x l, 3 x 3 or 5 x 5 and (2) depth of the output feature map - the number of filters computed by the convolution. Often these start with a depth of 32, continue to a depth of 64, and terminate with a depth of 128 or 256.
  • a convolution works by sliding these windows of size 3 x 3 or 5 x 5 over the 3D input feature map, stopping at every location, and extracting the 3D patch of surrounding features (shape (window_height, window_width, input_depth)).
  • Each such 3D patch is ten transformed (via a tensor product with the same learned weight matrix, called the convolution kernel) into a ID vector of shape (output_depth,). All of these vectors are then spatially reassembled into a 3D output map of shape (height, width, output_depth). Every spatial location in the output feature map corresponds to the same location in the input feature map (for example, the lower-right comer of the output contains information about the lower-right comer of the input).
  • the vector output [i, j, :] comes from the 3D patch input [i-1: i+1, j-l:J+l, :].
  • the full process is detailed in FIG. 4 (labeled as 400).
  • the convolutional neural network comprises convolution layers which perform the convolution operation between the input values and convolution filters (matrix of weights) that are learned over many gradient update iterations during the training.
  • (m, n) be the filter size and ITbe the matrix of weights
  • a convolution layer performs a convolution of the W with the input X by calculating the dot product W ⁇ x + b, where x is an instance of X and b is the bias.
  • the step size by which the convolution filters slide across the input is called the stride, and the filter area ( m c n) is called the receptive field.
  • a same convolution filter is applied across different positions of the input, which reduces the number of weights learned. It also allows location invariant learning, i.e., if an important pattern exists in the input, the convolution filters learn it no matter where it is in the sequence.
  • FIG. 5 depicts a block diagram 500 of training a convolutional neural network in accordance with one implementation of the technology disclosed.
  • the convolutional neural network is adjusted or trained so that the input data leads to a specific output estimate.
  • the convolutional neural network is adjusted using back propagation based on a comparison of the output estimate and the ground truth until the output estimate progressively matches or approaches the ground truth.
  • the convolutional neural network is trained by adjusting the weights between the neurons based on the difference between the ground truth and the actual output. This is mathematically described as:
  • the training rule is defined as:
  • the intermediary step in the training includes generating a feature vector from the input data using the convolution layers.
  • the gradient with respect to the weights in each layer, starting at the output, is calculated. This is referred to as the backward pass, or going backwards.
  • the weights in the network are updated using a combination of the negative gradient and previous weights.
  • the convolutional neural network uses a stochastic gradient update algorithm (such as ADAM) that performs backward propagation of errors by means of gradient descent.
  • ADAM stochastic gradient update algorithm
  • sigmoid function based back propagation algorithm is described below:
  • h is the weighted sum computed by a neuron.
  • the sigmoid function has the following derivative:
  • the algorithm includes computing the activation of all neurons in the network, yielding an output for the forward pass.
  • the activation of neuron Ytl in the hidden layers is described as:
  • the weights of the output layer are updated as:
  • the weights of the hidden layers are updated using the learning rate a as:
  • the convolutional neural network uses a gradient descent optimization to compute the error across all the layers.
  • the loss function is defined as l for the cost of predicting when the target isy, i.e. I (y, y ).
  • (X is the learning rate. Also, the loss is computed as the average over a set of VI data pairs. The computation is terminated when the learning rate (X is small enough upon linear convergence. In other implementations, the gradient is calculated using only selected data pairs fed to a Nesterov’s accelerated gradient and an adaptive gradient to inject computation efficiency.
  • the convolutional neural network uses a stochastic gradient descent (SGD) to calculate the cost function.
  • OC is the learning rate; is the momentum; and t is the current weight state before updating.
  • the convergence speed of SGD is approximately unit) when the learning rate (X are reduced both fast and slow enough.
  • the convolutional neural network uses different loss functions such as Euclidean loss and softmax loss.
  • an Adam stochastic optimizer is used by the convolutional neural network.
  • One variation on the iterative balanced sampling described previously is selecting the entire elite training set in one or two cycles instead of twenty. There may be enough distinction, learned by semi-supervised training, between known benign training examples and reliably classified predicted pathogenic variants that just one or two training cycles, or three to five training cycles, may be sufficient to assemble the elite training set.
  • Deep neural networks are a type of artificial neural networks that use multiple nonlinear and complex transforming layers to successively model high-level features. Deep neural networks provide feedback via backpropagation which carries the difference between observed and predicted output to adjust parameters. Deep neural networks have evolved with the availability of large training datasets, the power of parallel and distributed computing, and sophisticated training algorithms. Deep neural networks have facilitated major advances in numerous domains such as computer vision, speech recognition, and natural language processing.
  • Convolutional neural networks and recurrent neural networks (RNNs) are components of deep neural networks.
  • Convolutional neural networks have succeeded particularly in image recognition with an architecture that comprises convolution layers, nonlinear layers, and pooling layers.
  • Recurrent neural networks are designed to utilize sequential information of input data with cyclic connections among building blocks like perceptrons, long short-term memory units, and gated recurrent units.
  • many other emergent deep neural networks have been proposed for limited contexts, such as deep spatio-temporal neural networks, multi-dimensional recurrent neural networks, and convolutional auto-encoders.
  • the goal of training deep neural networks is optimization of the weight parameters in each layer, which gradually combines simpler features into complex features so that the most suitable hierarchical representations can be learned from data.
  • a single cycle of the optimization process is organized as follows. First, given a training dataset, the forward pass sequentially computes the output in each layer and propagates the function signals forward through the network. In the final output layer, an objective loss function measures error between the inferenced outputs and the given labels. To minimize the training error, the backward pass uses the chain rule to backpropagate error signals and compute gradients with respect to all weights throughout the neural network. Finally, the weight parameters are updated using optimization algorithms based on stochastic gradient descent.
  • stochastic gradient descent provides stochastic approximations by performing the updates for each small set of data examples.
  • optimization algorithms stem from stochastic gradient descent.
  • the Adagrad and Adam training algorithms perform stochastic gradient descent while adaptively modifying learning rates based on update frequency and moments of the gradients for each parameter, respectively.
  • regularization refers to strategies intended to avoid overfitting and thus achieve good generalization performance.
  • weight decay adds a penalty term to the objective loss function so that weight parameters converge to smaller absolute values.
  • Dropout randomly removes hidden units from neural networks during training and can be considered an ensemble of possible subnetworks.
  • maxout a new activation function
  • mnDrop a variant of dropout for recurrent neural networks
  • batch normalization provides a new regularization method through normalization of scalar features for each activation within a mini-batch and learning each mean and variance as parameters.
  • Convolutional neural networks have been adapted to solve sequence-based problems in genomics such as motif discovery, pathogenic variant identification, and gene expression inference.
  • Convolutional neural networks use a weightsharing strategy that is especially useful for studying DNA because it can capture sequence motifs, which are short, recurring local patterns in DNA that are presumed to have significant biological functions.
  • a hallmark of convolutional neural networks is the use of convolution filters. Unlike traditional classification approaches that are based on elaborately -designed and manually -crafted features, convolution filters perform adaptive learning of features, analogous to a process of mapping raw input data to the informative representation of knowledge.
  • the convolution filters serve as a series of motif scanners, since a set of such filters is capable of recognizing relevant patterns in the input and updating themselves during the training procedure.
  • Recurrent neural networks can capture long-range dependencies in sequential data of varying lengths, such as protein or DNA sequences.
  • the human reference genome harbors more than 70 million potential protein-altering missense substitutions, the vast majority of which are rare mutations whose effects on human health have not been characterized. These variants of unknown significance present a challenge for genome interpretation in clinical applications, and are a roadblock to the long term adoption of sequencing for population-wide screening and individualized medicine.
  • missense synonymous ratios across the allele frequency spectrum indicates that a large fraction of missense variants with population frequency ⁇ 0.1% are mildly deleterious, that is, neither pathogenic enough to warrant immediate removal from the population, nor neutral enough to be allowed to exist at high allele frequencies, consistent with prior observations on more limited population data.
  • missense: synonymous ratio is largely constant across the allele frequency spectrum.
  • the high allele frequency of these variants in the chimpanzee population indicates that they have already been through the sieve of natural selection in chimpanzee, and their neutral impact on fitness in human populations provides compelling evidence that the selective pressures on missense variants are highly concordant in the two species. The lower
  • missense synonymous ratio observed in chimpanzee is consistent with the larger effective population size in ancestral chimpanzee populations enabling more efficient filtering of mildly deleterious variants.
  • missense synonymous ratios are roughly equivalent across the allele frequency spectrum, other than a slight depletion of missense variation at high allele frequencies, which would be anticipated due to the inclusion of a small number of rare variants (—5-15%).
  • missense variation is markedly depleted at common allele frequencies compared to rare allele frequencies, especially at greater evolutionary distances, indicating that a substantial fraction of common missense variation in more distant species would experience negative selection in human populations. Nonetheless, the observation of a missense variant in more distant vertebrates still increases the likelihood of benign consequence, as the fraction of common missense variants depleted by natural selection is far less than the ⁇ 50% depletion for human missense variants at baseline.
  • Common primate variation also provides a clean validation dataset for evaluating existing methods that is completely independent of previously used training data, which has been hard to evaluate objectively because of the proliferation of meta-classifiers.
  • the accuracy of the deep learning classifier scales with the size of the training dataset, and variation data from each of the six primate species independently contributes to boosting the accuracy of the classifier.
  • the 504 known non-human primate species roughly 60% face extinction due to hunting and habitat loss, motivating urgency for a worldwide conservation effort that would benefit both these unique and irreplaceable species and our own.
  • a base refers to a nucleotide base or nucleotide, A (adenine), C (cytosine), T (thymine), or G (guanine).
  • chromosome refers to the heredity-bearing gene carrier of a living cell, which is derived from chromatin strands comprising DNA and protein components (especially histones). The conventional internationally recognized individual human genome chromosome numbering system is employed herein.
  • the term“site” refers to a unique position (e.g., chromosome ID, chromosome position and orientation) on a reference genome.
  • a site may be a residue, a sequence tag, or a segment’s position on a sequence.
  • locus may be used to refer to the specific location of a nucleic acid sequence or polymorphism on a reference chromosome.
  • sample refers to a sample, typically derived from a biological fluid, cell, tissue, organ, or organism containing a nucleic acid or a mixture of nucleic acids containing at least one nucleic acid sequence that is to be sequenced and/or phased.
  • samples include, but are not limited to sputum/oral fluid, amniotic fluid, blood, a blood fraction, fine needle biopsy samples (e.g., surgical biopsy, fine needle biopsy, etc.), urine, peritoneal fluid, pleural fluid, tissue explant, organ culture and any other tissue or cell preparation, or fraction or derivative thereof or isolated therefrom.
  • samples can be taken from any organism having chromosomes, including, but not limited to dogs, cats, horses, goats, sheep, cattle, pigs, etc.
  • the sample may be used directly as obtained from the biological source or following a pretreatment to modify the character of the sample.
  • pretreatment may include preparing plasma from blood, diluting viscous fluids and so forth.
  • Methods of pretreatment may also involve, but are not limited to, filtration, precipitation, dilution, distillation, mixing, centrifugation, freezing, lyophilization, concentration, amplification, nucleic acid fragmentation, inactivation of interfering components, the addition of reagents, lysing, etc.
  • sequence includes or represents a strand of nucleotides coupled to each other.
  • the nucleotides may be based on DNA or RNA. It should be understood that one sequence may include multiple subsequences. For example, a single sequence (e.g., of a PCR amplicon) may have 350 nucleotides.
  • the sample read may include multiple sub-sequences within these 350 nucleotides. For instance, the sample read may include first and second flanking subsequences having, for example, 20-50 nucleotides.
  • the first and second flanking subsequences may be located on either side of a repetitive segment having a corresponding sub-sequence (e.g., 40-100 nucleotides).
  • Each of the flanking sub-sequences may include (or include portions of) a primer sub-sequence (e.g., 10-30 nucleotides).
  • a primer sub-sequence e.g. 10-30 nucleotides.
  • the term“sub-sequence” will be referred to as“sequence,” but it is understood that two sequences are not necessarily separate from each other on a common strand.
  • the sequences may be given different labels (e.g., target sequence, primer sequence, flanking sequence, reference sequence, and the like). Other terms, such as“allele,” may be given different labels to differentiate between like objects.
  • paired-end sequencing refers to sequencing methods that sequence both ends of a target fragment. Paired-end sequencing may facilitate detection of genomic rearrangements and repetitive segments, as well as gene fusions and novel transcripts. Methodology for paired-end sequencing are described in PCT publication W007010252, PCT application Serial No. PCTGB2007/003798 and US patent application publication US 2009/0088327, each of which is incorporated by reference herein.
  • a series of operations may be performed as follows; (a) generate clusters of nucleic acids; (b) linearize the nucleic acids; (c) hybridize a first sequencing primer and carry out repeated cycles of extension, scanning and deblocking, as set forth above; (d) “invert” the target nucleic acids on the flow cell surface by synthesizing a complimentary copy; (e) linearize the resynthesized strand; and (f) hybridize a second sequencing primer and carry out repeated cycles of extension, scanning and deblocking, as set forth above.
  • the inversion operation can be carried out be delivering reagents as set forth above for a single cycle of bridge amplification.
  • reference genome refers to any particular known genome sequence, whether partial or complete, of any organism which may be used to reference identified sequences from a subject.
  • reference genome refers to any particular known genome sequence, whether partial or complete, of any organism which may be used to reference identified sequences from a subject.
  • A“genome” refers to the complete genetic information of an organism or virus, expressed in nucleic acid sequences.
  • a genome includes both the genes and the noncoding sequences of the DNA.
  • the reference sequence may be larger than the reads that are aligned to it.
  • the reference genome sequence is that of a full length human genome.
  • the reference genome sequence is limited to a specific human chromosome such as chromosome 13.
  • a reference chromosome is a chromosome sequence from human genome version hgl9. Such sequences may be referred to as chromosome reference sequences, although the term reference genome is intended to cover such sequences.
  • reference sequences include genomes of other species, as well as chromosomes, sub- chromosomal regions (such as strands), etc., of any species.
  • the reference genome is a consensus sequence or other combination derived from multiple individuals. However, in certain applications, the reference sequence may be taken from a particular individual.
  • the term“read” refers to a collection of sequence data that describes a fragment of a nucleotide sample or reference.
  • the term“read” may refer to a sample read and/or a reference read.
  • a read represents a short sequence of contiguous base pairs in the sample or reference.
  • the read may be represented symbolically by the base pair sequence (in ATCG) of the sample or reference fragment. It may be stored in a memory device and processed as appropriate to determine whether the read matches a reference sequence or meets other criteria.
  • a read may be obtained directly from a sequencing apparatus or indirectly from stored sequence information concerning the sample.
  • a read is a DNA sequence of sufficient length (e.g., at least about 25 bp) that can be used to identify a larger sequence or region, e.g., that can be aligned and specifically assigned to a chromosome or genomic region or gene.
  • Next-generation sequencing methods include, for example, sequencing by synthesis technology (Illumina), pyro sequencing (454), ion semiconductor technology (Ion Torrent sequencing), single-molecule realtime sequencing ( Pacific Biosciences) and sequencing by ligation (SOLiD sequencing).
  • the length of each read may vary from about 30 bp to more than 10,000 bp.
  • Illumina sequencing method using SOLiD sequencer generates nucleic acid reads of about 50 bp.
  • Ion Torrent Sequencing generates nucleic acid reads of up to 400 bp and 454 pyro sequencing generates nucleic acid reads of about 700 bp.
  • single-molecule real-time sequencing methods may generate reads of 10,000 bp to 15,000 bp. Therefore, in certain implementations, the nucleic acid sequence reads have a length of 30-100 bp, 50-200 bp, or 50-400 bp.
  • sample read refers to sequence data for a genomic sequence of interest from a sample.
  • the sample read comprises sequence data from a PCR amplicon having a forward and reverse primer sequence.
  • the sequence data can be obtained from any select sequence methodology.
  • the sample read can be, for example, from a sequencing-by-synthesis (SBS) reaction, a sequencing-by -ligation reaction, or any other suitable sequencing methodology for which it is desired to determine the length and/or identity of a repetitive element.
  • SBS sequencing-by-synthesis
  • the sample read can be a consensus (e.g., averaged or weighted) sequence derived from multiple sample reads.
  • providing a reference sequence comprises identifying a locus-of-interest based upon the primer sequence of the PCR amplicon.
  • the term“raw fragment” refers to sequence data for a portion of a genomic sequence of interest that at least partially overlaps a designated position or secondary position of interest within a sample read or sample fragment.
  • Non-limiting examples of raw fragments include a duplex stitched fragment, a simplex stitched fragment, a duplex un-stitched fragment and a simplex un-stitched fragment.
  • the term“raw” is used to indicate that the raw fragment includes sequence data having some relation to the sequence data in a sample read, regardless of whether the raw fragment exhibits a supporting variant that corresponds to and authenticates or confirms a potential variant in a sample read.
  • the term“raw fragment” does not indicate that the fragment necessarily includes a supporting variant that validates a variant call in a sample read. For example, when a sample read is determined by a variant call application to exhibit a first variant, the variant call application may determine that one or more raw fragments lack a corresponding type of“supporting” variant that may otherwise be expected to occur given the variant in the sample
  • the terms“mapping”,“aligned,”“alignment,” or“aligning” refer to the process of comparing a read or tag to a reference sequence and thereby determining whether the reference sequence contains the read sequence. If the reference sequence contains the read, the read may be mapped to the reference sequence or, in certain implementations, to a particular location in the reference sequence. In some cases, alignment simply tells whether or not a read is a member of a particular reference sequence (i.e., whether the read is present or absent in the reference sequence). For example, the alignment of a read to the reference sequence for human chromosome 13 will tell whether the read is present in the reference sequence for chromosome 13. A tool that provides this information may be called a set membership tester.
  • an alignment additionally indicates a location in the reference sequence where the read or tag maps to. For example, if the reference sequence is the whole human genome sequence, an alignment may indicate that a read is present on chromosome 13, and may further indicate that the read is on a particular strand and/or site of chromosome 13.
  • micro-indel refers to the insertion and/or the deletion of bases in the DNA of an organism.
  • a micro-indel represents an indel that results in a net change of 1 to 50 nucleotides. In coding regions of the genome, unless the length of an indel is a multiple of 3, it will produce a frameshift mutation.
  • Indels can be contrasted with point mutations. An indel inserts and deletes nucleotides from a sequence, while a point mutation is a form of substitution that replaces one of the nucleotides without changing the overall number in the DNA. Indels can also be contrasted with a Tandem Base Mutation (TBM), which may be defined as substitution at adjacent nucleotides (primarily substitutions at two adjacent nucleotides, but substitutions at three adjacent nucleotides have been observed).
  • TBM Tandem Base Mutation
  • variant refers to a nucleic acid sequence that is different from a nucleic acid reference.
  • Typical nucleic acid sequence variants include without limitation single nucleotide polymorphism (SNP), short deletion and insertion polymorphisms (Indel), copy number variation (CNV), micro satellite markers or short tandem repeats and structural variation.
  • Somatic variant calling is the effort to identify variants present at low frequency in the DNA sample. Somatic variant calling is of interest in the context of cancer treatment. Cancer is caused by an accumulation of mutations in DNA. A DNA sample from a tumor is generally heterogeneous, including some normal cells, some cells at an early stage of cancer progression (with fewer mutations), and some late-stage cells (with more mutations).
  • somatic mutations will often appear at a low frequency. For example, a SNV might be seen in only 10% of the reads covering a given base.
  • a variant that is to be classified as somatic or germline by the variant classifier is also referred to herein as the“variant under test”.
  • noise refers to a mistaken variant call resulting from one or more errors in the sequencing process and/or in the variant call application.
  • variable frequency represents the relative frequency of an allele (variant of a gene) at a particular locus in a population, expressed as a fraction or percentage.
  • the fraction or percentage may be the fraction of all chromosomes in the population that carry that allele.
  • sample variant frequency represents the relative frequency of an allele/variant at a particular locus/position along a genomic sequence of interest over a“population” corresponding to the number of reads and/or samples obtained for the genomic sequence of interest from an individual.
  • a baseline variant frequency represents the relative frequency of an allele/variant at a particular locus/position along one or more baseline genomic sequences where the“population” corresponding to the number of reads and/or samples obtained for the one or more baseline genomic sequences from a population of normal individuals.
  • VAF variant allele frequency
  • the terms“position”,“designated position”, and“locus” refer to a location or coordinate of one or more nucleotides within a sequence of nucleotides.
  • the terms“position”,“designated position”, and“locus” also refer to a location or coordinate of one or more base pairs in a sequence of nucleotides.
  • haplotype refers to a combination of alleles at adjacent sites on a chromosome that are inherited together.
  • a haplotype may be one locus, several loci, or an entire chromosome depending on the number of recombination events that have occurred between a given set of loci, if any occurred.
  • threshold refers to a numeric or non-numeric value that is used as a cutoff to characterize a sample, a nucleic acid, or portion thereof (e.g., a read).
  • a threshold may be varied based upon empirical analysis. The threshold may be compared to a measured or calculated value to determine whether the source giving rise to such value suggests should be classified in a particular manner. Threshold values can be identified empirically or analytically. The choice of a threshold is dependent on the level of confidence that the user wishes to have to make the classification. The threshold may be chosen for a particular purpose (e.g., to balance sensitivity and selectivity).
  • the term“threshold” indicates a point at which a course of analysis may be changed and/or a point at which an action may be triggered.
  • a threshold is not required to be a predetermined number. Instead, the threshold may be, for instance, a function that is based on a plurality of factors. The threshold may be adaptive to the circumstances. Moreover, a threshold may indicate an upper limit, a lower limit, or a range between limits.
  • a metric or score that is based on sequencing data may be compared to the threshold.
  • the terms“metric” or“score” may include values or results that were determined from the sequencing data or may include functions that are based on the values or results that were determined from the sequencing data.
  • the metric or score may be adaptive to the circumstances. For instance, the metric or score may be a normalized value.
  • one or more implementations may use count scores when analyzing the data. A count score may be based on the number of sample reads. The sample reads may have undergone one or more filtering stages such that the sample reads have at least one common characteristic or quality.
  • each of the sample reads that are used to determine a count score may have been aligned with a reference sequence or may be assigned as a potential allele.
  • the number of sample reads having a common characteristic may be counted to determine a read count.
  • Count scores may be based on the read count. In some implementations, the count score may be a value that is equal to the read count. In other implementations, the count score may be based on the read count and other information. For example, a count score may be based on the read count for a particular allele of a genetic locus and a total number of reads for the genetic locus. In some implementations, the count score may be based on the read count and previously -obtained data for the genetic locus.
  • the count scores may be normalized scores between predetermined values.
  • the count score may also be a function of read counts from other loci of a sample or a function of read counts from other samples that were concurrently ran with the sample-of-interest.
  • the count score may be a function of the read count of a particular allele and the read counts of other loci in the sample and/or the read counts from other samples.
  • the read counts from other loci and/or the read counts from other samples may be used to normalize the count score for the particular allele.
  • the terms“coverage” or“fragment coverage” refer to a count or other measure of a number of sample reads for the same fragment of a sequence.
  • a read count may represent a count of the number of reads that cover a corresponding fragment.
  • the coverage may be determined by multiplying the read count by a designated factor that is based on historical knowledge, knowledge of the sample, knowledge of the locus, etc.
  • read depth refers to the number of sequenced reads with overlapping alignment at the target position. This is often expressed as an average or percentage exceeding a cutoff over a set of intervals (such as exons, genes, or panels). For example, a clinical report might say that a panel average coverage is 1, 105 c with 98% of targeted bases covered >100 c .
  • base call quality score or“Q score” refer to a PHRED-scaled probability ranging from 0- 20 inversely proportional to the probability that a single sequenced base is correct. For example, a T base call with Q of 20 is considered likely correct with a confidence P-value of 0.01. Any base call with Q ⁇ 20 should be considered low quality, and any variant identified where a substantial proportion of sequenced reads supporting the variant are of low quality should be considered potentially false positive.
  • variant reads or“variant read numbed’ refer to the number of sequenced reads supporting the presence of the variant.
  • nucleic acid sequences may be analyzed in accordance with the methods and systems described in US Patent Application Publication No. 2016/0085910 and US Patent Application Publication No. 2013/0296175, the complete subject matter of which are expressly incorporated by reference herein in their entirety.
  • a sequencing process includes receiving a sample that includes or is suspected of including nucleic acids, such as DNA.
  • the sample may be from a known or unknown source, such as an animal (e.g., human), plant, bacteria, or fungus.
  • the sample may be taken directly from the source. For instance, blood or saliva may be taken directly from an individual. Alternatively, the sample may not be obtained directly from the source.
  • one or more processors direct the system to prepare the sample for sequencing.
  • the preparation may include removing extraneous material and/or isolating certain material (e.g., DNA).
  • the biological sample may be prepared to include features for a particular assay. For example, the biological sample may be prepared for sequencing-by-synthesis (SBS).
  • SBS sequencing-by-synthesis
  • the preparing may include amplification of certain regions of a genome.
  • the preparing may include amplifying predetermined genetic loci that are known to include STRs and/or SNPs.
  • the genetic loci may be amplified using predetermined primer sequences.
  • the one or more processors may direct the system to sequence the sample.
  • the sequencing may be performed through a variety of known sequencing protocols.
  • the sequencing includes SBS.
  • SBS a plurality of fluorescently-labeled nucleotides are used to sequence a plurality of clusters of amplified DNA (possibly millions of clusters) present on the surface of an optical substrate (e.g., a surface that at least partially defines a channel in a flow cell).
  • the flow cells may contain nucleic acid samples for sequencing where the flow cells are placed within the appropriate flow cell holders.
  • the nucleic acids can be prepared such that they comprise a known primer sequence that is adjacent to an unknown target sequence.
  • one or more differently labeled nucleotides, and DNA polymerase, etc. can be flowed into/through the flow cell by a fluid flow subsystem.
  • Either a single type of nucleotide can be added at a time, or the nucleotides used in the sequencing procedure can be specially designed to possess a reversible termination property, thus allowing each cycle of the sequencing reaction to occur simultaneously in the presence of several types of labeled nucleotides (e.g., A, C, T, G).
  • the nucleotides can include detectable label moieties such as fluorophores.
  • Non-incorporated nucleotides can be washed away by flowing a wash solution through the flow cell.
  • One or more lasers may excite the nucleic acids and induce fluorescence.
  • the fluorescence emitted from the nucleic acids is based upon the fluorophores of the incorporated base, and different fluorophores may emit different wavelengths of emission light.
  • a deblocking reagent can be added to the flow cell to remove reversible terminator groups from the DNA strands that were extended and detected. The deblocking reagent can then be washed away by flowing a wash solution through the flow cell.
  • the flow cell is then ready for a further cycle of sequencing starting with introduction of a labeled nucleotide as set forth above.
  • the fluidic and detection operations can be repeated several times to complete a sequencing run.
  • Example sequencing methods are described, for example, in Bentley et al, Nature 456:53-59 (2008), International Publication No. WO 04/018497; U.S. Pat. No. 7,057,026; International Publication No. WO 91/06678; International Publication No. WO 07/123744; U.S. Pat. No. 7,329,492; U.S. Patent No. 7,211,414; U.S. Patent No. 7,315,019; U.S. Patent No. 7,405,281, and U.S. Patent Application Publication No. 2008/0108082, each of which is incorporated herein by reference.
  • nucleic acids can be attached to a surface and amplified prior to or during sequencing.
  • amplification can be carried out using bridge amplification to form nucleic acid clusters on a surface.
  • Useful bridge amplification methods are described, for example, in U.S. Patent No. 5,641,658; U.S. Patent Application Publication No. 2002/0055100; U.S. Patent No. 7,115,400; U.S. Patent Application Publication No. 2004/0096853; U.S. Patent Application Publication No. 2004/0002090; U.S. Patent Application Publication No. 2007/0128624; and U.S. Patent Application Publication No.
  • RCA rolling circle amplification
  • One example SBS protocol exploits modified nucleotides having removable 3’ blocks, for example, as described in International Publication No. WO 04/018497, U.S. Patent Application Publication No.
  • repeated cycles of SBS reagents can be delivered to a flow cell having target nucleic acids attached thereto, for example, as a result of the bridge amplification protocol.
  • the nucleic acid clusters can be converted to single stranded form using a linearization solution.
  • the linearization solution can contain, for example, a restriction endonuclease capable of cleaving one strand of each cluster.
  • cleavage can be used as an alternative to restriction enzymes or nicking enzymes, including inter alia chemical cleavage (e.g., cleavage of a diol linkage with periodate), cleavage of abasic sites by cleavage with endonuclease (for example‘USER’, as supplied by NEB, Ipswich, Mass., USA, part number M5505 S), by exposure to heat or alkali, cleavage of ribonucleotides incorporated into amplification products otherwise comprised of deoxyribonucleotides, photochemical cleavage or cleavage of a peptide linker.
  • a sequencing primer can be delivered to the flow cell under conditions for hybridization of the sequencing primer to the target nucleic acids that are to be sequenced.
  • a flow cell can then be contacted with an SBS extension reagent having modified nucleotides with removable 3’ blocks and fluorescent labels under conditions to extend a primer hybridized to each target nucleic acid by a single nucleotide addition. Only a single nucleotide is added to each primer because once the modified nucleotide has been incorporated into the growing polynucleotide chain complementary to the region of the template being sequenced there is no free 3’-OH group available to direct further sequence extension and therefore the polymerase cannot add further nucleotides.
  • the SBS extension reagent can be removed and replaced with scan reagent containing components that protect the sample under excitation with radiation. Example components for scan reagent are described in U.S. Patent Application Publication No.
  • the extended nucleic acids can then be fluorescently detected in the presence of scan reagent. Once the fluorescence has been detected, the 3’ block may be removed using a deblock reagent that is appropriate to the blocking group used.
  • Example deblock reagents that are useful for respective blocking groups are described in W0004018497, US 2007/0166705A1 and U.S. Patent No. 7,057,026, each of which is incorporated herein by reference.
  • the deblock reagent can be washed away leaving target nucleic acids hybridized to extended primers having 3’-OH groups that are now competent for addition of a further nucleotide.
  • each cycle can include separate operations of extension reagent delivery followed by separate operations of scan reagent delivery and detection, in which case two or more of the nucleotides can have the same label and can be distinguished based on the known order of delivery.
  • the sequencing data may be formatted in various manners, such as in a .BAM file.
  • the sequencing data may include, for example, a number of sample reads.
  • the sequencing data may include a plurality of sample reads that have corresponding sample sequences of the nucleotides. Although only one sample read is discussed, it should be understood that the sequencing data may include, for example, hundreds, thousands, hundreds of thousands, or millions of sample reads. Different sample reads may have different numbers of nucleotides.
  • a sample read may range between 10 nucleotides to about 500 nucleotides or more.
  • the sample reads may span the entire genome of the source(s).
  • the sample reads are directed toward predetermined genetic loci, such as those genetic loci having suspected STRs or suspected SNPs.
  • Each sample read may include a sequence of nucleotides, which may be referred to as a sample sequence, sample fragment or a target sequence.
  • the sample sequence may include, for example, primer sequences, flanking sequences, and a target sequence.
  • the number of nucleotides within the sample sequence may include 30, 40, 50, 60, 70, 80, 90, 100 or more.
  • one or more of the sample reads (or sample sequences) includes at least 150 nucleotides, 200 nucleotides, 300 nucleotides, 400 nucleotides, 500 nucleotides, or more.
  • the sample reads may include more than 1000 nucleotides, 2000 nucleotides, or more.
  • the sample reads (or the sample sequences) may include primer sequences at one or both ends.
  • the one or more processors analyze the sequencing data to obtain potential variant call(s) and a sample variant frequency of the sample variant call(s).
  • the operation may also be referred to as a variant call application or variant caller.
  • the variant caller identifies or detects variants and the variant classifier classifies the detected variants as somatic or germline.
  • Alternative variant callers may be utilized in accordance with implementations herein, wherein different variant callers may be used based on the type of sequencing operation being performed, based on features of the sample that are of interest and the like.
  • a variant call application such as the PiscesTM application by Illumina Inc. (San Diego, CA) hosted at
  • chimpanzees comprise the next closest extant species, and share 99.4% amino acid sequence identity.
  • the near-identity of the protein-coding sequence in humans and chimpanzees suggests that purifying selection operating on chimpanzee protein-coding variants might also model the consequences on fitness of human mutations that are identical -by -state.
  • the pathogenicity prediction network takes as input the 51 -length amino acid sequence centered at the variant of interest, and the outputs of the secondary structure and solvent accessibility networks (FIG. 2 and FIG. 3) with the missense variant substituted in at the central position.
  • Three 51-length position frequency matrices are generated from multiple sequence alignments of 99 vertebrates, including one for 11 primates, one for 50 mammals excluding primates, and one for 38 vertebrates excluding primates and mammals.
  • the secondary structure deep learning network predicts a three-state secondary structure at each amino acid position: alpha helix (H), beta sheet (B), and coils (C).
  • the solvent accessibility network predicts a three-state solvent accessibility at each amino acid position: buried (B), intermediate (I), and exposed (E). Both networks can only take the flanking amino acid sequence as their inputs, and can be trained using labels from known non- redundant crystal structures in the Protein DataBank.
  • a single length position frequency matrix which is generated from the multiple sequence alignments for all 99 vertebrates, also with length 51 and depth 20, can be used.
  • the final two layers for the secondary structure and solvent models can be removed and the output of the network can be directly connected to the input of the pathogenicity model.
  • Exemplary testing accuracy achieved for the three-state secondary structure prediction model was 79.86%. There was no substantial difference when comparing the predictions of the neural network when using DSSP-annotated structure labels for the approximately ⁇ 4,000 human proteins that had crystal structures versus using predicted structure labels only.
  • FIG. 2 shows an example architecture 200 of a deep residual network for pathogenicity prediction, referred to herein as“PrimateAI”.
  • ID refers to 1-dimensional convolutional layer.
  • Predicted pathogenicity is on a scale from 0 (benign) to 1 (pathogenic).
  • the network takes as input the human amino acid (AA) reference and alternate sequence (51 AAs) centered at the variant, the position weight matrix (PWM) conservation profiles calculated from 99 vertebrate species, and the outputs of secondary structure and solvent accessibility prediction deep learning networks, which predict three-state protein secondary structure (helix— H, beta sheet— B, and coil— C) and three-state solvent accessibility (buried— B, intermediate— I, and exposed— E).
  • AA human amino acid
  • PWM position weight matrix
  • FIG. 3 depicts a schematic illustration 300 of PrimateAI, the deep learning network architecture for pathogenicity classification.
  • the inputs to the model include 51 amino acids (AA) of flanking sequence for both the reference sequence and the sequence with the variant substituted in, conservation represented by three 51-AA -length position-weighted matrices from primate, mammal, and vertebrate alignments, and the outputs of pre-trained secondary structure network and solvent accessibility network (also 51 AA in length).
  • AA amino acids
  • FIG. 1 shows an architectural level schematic 100 of the system in accordance with an implementation. Because FIG. 1 is an architectural diagram, certain details are intentionally omitted to improve the clarity of description. The discussion of FIG. 1 is organized as follows. First, the elements of the figure are described, followed by their interconnection. Then, the use of the elements in the system is described in greater detail.
  • the system includes four training datasets: pathogenic missense training examples 121, supplemental benign training examples 131, benign missense training examples 161 and supplemental benign training examples 181.
  • the system further includes a trainer 114, a tester 116, a position frequency matrix (PFM) calculator 184, an input encoder 186, a variant pathogenicity prediction model 157, and network(s) 155.
  • the supplemental benign training examples 131 correspond to the pathogenic missense training examples 121 and are therefore placed together in a box with broken lines.
  • supplemental benign training examples 181 correspond to the benign missense training examples 161 and therefore, both datasets are shown in the same box.
  • the system is described with PrimateAI as an example variant pathogenicity prediction model 157 which takes as input amino acid sequences flanking the variant of interest and orthologous sequence alignments in other species.
  • the detailed architecture of the PrimateAI model for pathogenicity prediction is presented above in reference to FIG. 3.
  • the input of the amino acid sequence includes the variant of interest.
  • the term“variant” refers to an amino acid sequence that is different from an amino acid reference sequence.
  • a tri-nucleotide base sequence also referred to as a codon
  • the codons“AAA” and“AAG” represent Lysine amino acid (also referred to by a symbol“K”).
  • An amino acid sequence variant can be caused by a single nucleotide polymorphism (SNP).
  • SNP is a variation in a single nucleotide that occurs at a specific locus in the gene and is observed to some appreciable degree within a population (e.g., >1%).
  • exons There are two types of SNPs: a synonymous SNP and a missense SNP.
  • a synonymous SNP is a type of protein-coding SNP that changes a first codon for an amino acid into a second codon for the same amino acid.
  • a missense SNP on the other hand includes change of a first codon for a first amino acid into a second codon for a second amino acid.
  • FIG. 6 presents an example 600 of“protein sequence pairs” for a missense variant and a corresponding constructed synonymous variant.
  • the phrase“protein sequence pair” or simply a“sequence paid’ refers to a reference protein sequence and an alternative protein sequence.
  • the reference protein sequence comprises reference amino acids expressed by reference codons or tri-nucleotide bases.
  • the alternative protein sequence comprises alternative amino acids expressed by alternative codons or tri-nucleotide bases such that the alternative protein sequence results due to a variant occurring in the reference codon that expresses the reference amino acids of the reference protein sequence.
  • supplemental benign training example a supplemental benign synonymous counterpart training example (referred above as supplemental benign training example) corresponding to a missense variant.
  • the missense variant can be a pathogenic missense training example or a benign missense training example.
  • a SNP occurs in the same chromosome at position 6 resulting in an alternate sequence with codon“TCT” at the same positions i.e., 5 :7.
  • FIG. 6 only shows amino acids and corresponding codons in the sequence pairs at target location. Flanking amino acids and respective codons in the sequence pairs are not shown.
  • the missense variant is labeled as pathogenic (labeled as“1”).
  • the technology disclosed constructs a counterpart supplemental benign training example for the corresponding missense variant.
  • the reference sequence in the sequence pair for the constructed supplemental benign training example is the same as the reference sequence in the missense variant shown in the left part of FIG. 6.
  • the right part of FIG. 6 shows a supplemental benign training example which is a synonymous counterpart with the same reference sequence codon“ccc” i n chromosome 1 at positions 5 :7 as in the reference sequence for missense variant.
  • the alternative sequence constructed for synonymous counterpart has a SNP at position number 7 which results in a codon“TTC”.
  • This codon results in the same amino acid Phenylalanine (F) in the alternate sequence as in the reference sequence at the same position in the same chromosome.
  • Two different codons in the same chromosome at the same positions express the same amino acid therefore, the synonymous counterpart is labeled as benign (or labeled as“0”).
  • the two different codons in the same position in the reference and the alternative sequence express the same amino acid at the target location.
  • the benign counterpart is not randomly constructed; instead it is selected from synonymous variants observed in a sequenced population.
  • the technology disclosed constructs supplemental benign training examples to contrast the pathogenic missense training example to reduce overfitting of the variant pathogenicity prediction model during training.
  • a supplemental benign training example need not be synonymous.
  • the technology disclosed can also construct supplemental benign training examples having the same amino acid in the alternate sequence that are constructed by identical tri-nucleotide codons as in the reference sequence.
  • the associated position frequency matrix (PFM) is the same for identical amino acid sequences, regardless of whether the amino acids are expressed by synonymous or identical codons. Therefore, such supplemental training examples have the same effect of reducing the overfitting of the variant pathogenicity prediction model during training as synonymous counterpart training example presented in FIG. 6.
  • the trainer 114 uses the four training datasets presented in FIG. 1 to train the variant pathogenicity prediction model.
  • the variant pathogenicity prediction model is implemented as a convolutional neural network (CNN).
  • CNN convolutional neural network
  • the training of CNN is described above with reference to FIG. 5.
  • Training includes adjusting the CNN using back propagation based on a comparison of the output estimate and the ground truth until the output estimate progressively matches or approaches the ground truth.
  • the tester 116 uses test dataset to benchmark the variant pathogenicity prediction model.
  • the input encoder 186 converts categorical inputs data such as reference and alternative amino acid sequences into a form that can be provided as input to the variant pathogenicity prediction model. This is further explained using example reference and alternative sequences in FIG. 13.
  • the PFM Calculator 184 calculates the position frequency matrix (PFM) which is also referred to as position-specific scoring matrix (PSSM) or position weight matrix (PWM).
  • PFM position frequency matrix
  • PSSM position-specific scoring matrix
  • PWM position weight matrix
  • the technology disclosed calculates three PFMs, one each for primates, mammals and vertebrates.
  • the length of amino acid sequences for each of the three PFMs can be 51 with a target amino acid flanked upstream and downstream by at least 25 amino acids.
  • the PFMs have 20 rows for amino acids and 51 columns for positions of amino acids in the amino acid sequences.
  • the PFM calculator calculates the first PFM with amino acid sequences for 11 primates, the second PFM with amino acid sequences for 48 mammals and a third PFM with amino acid sequences for 40 vertebrates.
  • a cell in a PFM is a count of the occurrences of an amino acid at a specific position in the sequence.
  • the amino acid sequences for the three PFMs are aligned.
  • results of position-wise calculation for primate, mammal and vertebrate PFMs for each amino acid position in the reference amino acid sequence or the alternative amino acid sequence are stored on the position-wise or ordinal position-basis in the same order as the amino acid positions occur in the reference amino acid sequence or the alternative amino acid sequence.
  • FIG. 7 presents an illustration 700 of pre-training epochs 1 to 5 in which approximately 400,000 benign supplemental training examples 131 are combined with approximately 400,000 pathogenic variants 121 predicted from the deep learning models. Fewer benign supplemental training examples such as approximately 100,000, 200,000 or 300,000 can be combined with pathogenic variants.
  • the pathogenic variant dataset is generated in 20 cycles using random samples from approximately 68 million synthetic variants as described above. In another implementation, the pathogenic variant dataset can be generated in one cycle from approximately 68 million synthetic variants.
  • the pathogenic variants 121 and supplemental benign training examples 131 are given as input to an ensemble of networks in the first five epochs.
  • approximately 400,000 supplemental benign training examples 181 are combined with approximately 400,000 benign variants 161 for ensemble training during pretraining epochs. Fewer benign training examples such as approximately 100,000, 200,000 or 300,000 can be combined with benign variants.
  • the supplemental benign datasets 131 and 181 are not given as input for the rest of the training epochs 6 to n as shown in example 800 in FIG. 8.
  • the training of the ensemble of networks continues over multiple epochs with pathogenic variants dataset and benign variants dataset.
  • the training is terminated after a predetermined number of training epochs or when a termination condition is reached.
  • the trained network is used during inference to evaluate synthetic variants 810 as shown in example 900 in FIG. 9.
  • the trained network predicts a variant as pathogenic or benign.
  • a PFM for an example supplemental benign training example 1012 which is constructed as a counterpart of a pathogenic missense variant training example 1002, illustrated in FIG. 10 (referred to by a numeral 1000).
  • a PFM is generated or referenced for a training example.
  • the PFM for the training example depends only on the position of the reference sequence, so both training examples 1002 and 1012 have the same PFM.
  • the first training example 1002 is a pathogenic/unlabeled variant.
  • the second training example 1012 is a counterpart supplemental benign training example corresponding to the training example 1002.
  • the training example 1002 has a reference sequence 1002R and an alternative sequence 1002A.
  • a first PFM is accessed or generated for the training example 1002 based only on the position of reference sequence 1002R.
  • Training example 1012 has a reference sequence 1012R and alternative sequence 1012A.
  • the first PFM for example 1002 can be reused for example 1012.
  • the PFM is calculated using amino acid sequences from multiple species, such as 99 species of primates, mammals, and vertebrates, as an indication of conservation of sequences across species. Humans may or may not be among the species represented in calculation of the PFM.
  • the cells in this PFM include counts of occurrences of amino acids, across species, in the sequences.
  • PFM 1022 is a starting point for a PFM, which illustrates one hot encoding of a single sequence in a training example. When a PFM is complete, for the example of 99 species, positions that are completely conserved across species will have a value of“99” instead of“ 1”. Partial conservation will result in two or more rows in a column having values that sum to 99, in this example.
  • the reference and alternative sequences both have the same PFM, because the PFM depends on the overall sequence position, not on the amino acid in the center position of the sequence.
  • the example reference and alternative amino acid sequences for both pathogenic/unlabeled training example 1002 and supplemental benign training example 1012 as shown in FIG. 10 have 51 amino acids.
  • the reference amino acid sequence 1002R has Arginine amino acid represented by“R” at position 26 (also referred to as target position) in the sequence.
  • R Arginine amino acid
  • one of the six tri-nucleotide bases or codons expresses amino acid“R”.
  • codons See FIG. 10 for example to simplify the illustration and rather focus on calculation of PFMs.
  • FIG. 11 presents an example 1100 in which two PFMs are calculated for an example benign missense variant 1102 and corresponding supplemental benign training example 1122.
  • reference sequences 1102R and 1112R are the same for both benign missense variant 1102 and supplemental benign training example 1112.
  • Their respective alternative sequences 1102A and 1112 A are also shown in the FIG. 11.
  • Two PFMs are generated or referenced for the two reference sequences as describe above for the example presented in FIG. 10. Both PFMs are same, only one PFM 1122 is shown in FIG. 11 for illustration purposes. Both of these PFMs represent benign (“0”) labeled amino acid sequences.
  • the technology disclosed calculates three PFMs one each for 11 primate sequences, 48 mammal sequences and 40 vertebrate sequences.
  • FIG. 12 illustrates the three PFMs 1218, 1228, and 1238 each with 20 rows and 51 columns.
  • the primate sequences do not include human reference sequence.
  • the primate sequences include the human reference sequence.
  • the values of cells in the three PFMs are calculated by counting occurrences of an amino acid (row label) that is present in all sequences for the PFM at a given position (column label). For example, if three primate sequences have an amino acid“K” at position 26, the value of the cell with row label“K” and column label“26” has a value of“3”.
  • One hot encoding is a process by which categorical variables are converted into a form that can be provided an input to a deep learning model.
  • a categorical value represents an alpha-numerical value for an entry in the dataset.
  • the reference and alternative amino acid sequences each have 51 amino acid characters arranged in a sequence.
  • An amino acid character“T” at position“1” in a sequence represents an amino acid Threonine at the first position in the sequence.
  • the amino acid sequences are encoded by assigning a value of“ 1” in the cell with row label“T” and column label“1” in the one hot encoded representation.
  • the one hot encoded representation for an amino acid sequence has 0s in the cells except the cells which represent amino acid (row label) occurring in a particular position (column label).
  • FIG. 13 illustrates an example 1300 in which reference and alternative sequence for a supplemental benign training example are represented as one hot encoded.
  • the reference and alternative amino acid sequences are given as input one hot encoded form to the variant pathogenicity prediction model.
  • FIG. 14 includes an illustration 1400 depicting the inputs to the variant pathogenicity prediction model.
  • the inputs include human reference and alternative amino acid sequences in one hot encoded form and the PFM 1218 for primates, the PFM 1228 for mammals and the PFM 1238 for vertebrates. As described above, the PFM for primates can include only non-human primates or human and non-human primates.
  • the performance of the neural network-based model improves by using the pre-training epochs presented above.
  • the following table presents example test results. The results in the table are organized under six headings. We briefly describe the headings before presenting the results. “Replicate” columns presents results for 20 replicate runs.
  • Each run can be an ensemble of eight models with different random seeds.
  • “Accuracy” is the proportion of withheld 10,000 primate benign variants that are classified as benign.
  • “Pvalue_DDD” presents results of Wilcoxon rank test to evaluate how well the separation of de novo mutations of affected children with developmental disorders from their unaffected siblings.
  • “pvalue_605genes” presents results of similar tests as pvalue_DDD, except in this case we used de novo mutations within the 605 disease related genes.
  • “Corr_RK_RW” presents the correlation of primateAI scores between amino acid change from R to K and from R to W. A smaller value of Corr_RK_RW indicates better performance.
  • “Pvalue_Corr” presents the pvalue of the correlation in the previous column, i.e., Corr_RK_RW.
  • Implementations that are not mutually exclusive are taught to be combinable. One or more features of an implementation can be combined with other implementations. This disclosure periodically reminds the user of these options. Omission from some implementations of recitations that repeat these options should not be taken as limiting the combinations taught in the preceding sections - these recitations are hereby incorporated forward by reference into each of the following implementations.
  • a system implementation of the technology disclosed includes one or more processors coupled to the memory.
  • the memory is loaded with computer instructions to reduce overfitting of a neural network-implemented model that processes sequences of amino acids and accompanying position frequency matrices (PFMs).
  • the system includes logic to generate benign labeled supplemental training example sequence pairs that include a start location, through a target amino acid location, to an end location.
  • a supplemental sequence pair matches the start location and the end location of a missense training example sequence pair. It has identical amino acids in a reference and an alternate sequence of amino acids.
  • the system includes logic to input with each supplemental sequence pair a supplemental training PFM that is identical to the PFM of the missense training example sequence pair at the matching start and end location.
  • the system includes logic to train the neural network-implemented model using the benign training example sequence pairs and the supplemental training example PFMs, and the missense training example sequence pairs and PFMs of the missense training example sequence pairs at the matching start and end locations.
  • the training influence of the training PFMs is attenuated during the training.
  • This system implementation and other systems disclosed optionally include one or more of the following features.
  • System can also include features described in connection with methods disclosed. In the interest of conciseness, alternative combinations of system features are not individually enumerated. Features applicable to systems, methods, and articles of manufacture are not repeated for each statutory class set of base features. The reader will understand how features identified in this section can readily be combined with base features in other statutory classes.
  • the system can include logic to construct supplemental sequence pairs such that each supplemental sequence pair matches the start location and the end location of a benign missense training example sequence pair.
  • the system can include logic to construct supplemental sequence pairs such that each supplemental sequence pair matches the start location and the end location of a pathogenic missense training example sequence pair.
  • the system includes logic to modify the training of the neural network-implemented model to cease using the supplemental training example sequence pairs and the supplemental training PFMs after a predetermined number of training epochs.
  • the system includes logic to modify the training of the neural network-implemented model to cease using the supplemental training example sequence pairs and the supplemental training PFMs after three training epochs.
  • the system includes logic to modify the training of the neural network-implemented model to cease using the supplemental training example sequence pairs and the supplemental training PFMs after five training epochs.
  • the ratio of the supplemental training example sequence pairs to the pathogenic training example sequence pairs can be between 1: 1 and 1:8.
  • the system can use different values for the ranges for example, between 1: 1 and 1: 12, between 1: 1 and 1: 16 and between 1: 1 and 1:24.
  • the ratio of the supplemental training example sequence pairs to the benign training example sequence pairs can be between 1:2 and 1:8.
  • the system can use different values for the ranges for example, between 1: 1 and 1: 12, between 1: 1 and 1: 16 and between 1: 1 and 1:24.
  • the system includes logic to create the supplemental PFMs, amino acid locations from data for nonhuman primates and non-primate mammals.
  • implementations may include a non-transitory computer readable storage medium storing instructions executable by a processor to perform functions of the system described above.
  • implementations may include a method performing the functions of the system described above.
  • a method implementation of the technology disclosed includes generating benign labeled supplemental training example sequence pairs that include a start location, through a target amino acid location, to an end location. Each supplemental sequence pair matches the start location and the end location of a missense training example sequence pair. It has identical amino acids in a reference and an alternate sequence of amino acids.
  • the method includes inputting a supplemental training PFM with each supplemental sequence pair that is identical to the PFM of the missense training example sequence pair at the matching start and end location.
  • the method includes training the neural network-implemented model using the benign training example sequence pairs and the supplemental training example PFMs, and the missense training example sequence pairs, and the PFMs of the missense at the matching start and end locations. The training influence of the training PFMs is attenuated during the training.
  • This method implementation and other methods disclosed optionally include one or more of the following features. Methods can also include features described in connection with systems disclosed. The reader will understand how features identified in this section can readily be combined with base features in other statutory classes.
  • implementations may include a set of one or more non-transitory computer readable storage media collectively storing computer program instructions executable by one or more processors to reduce overfitting of a neural network-implemented model that processes sequences of amino acids and accompanying position frequency matrices (PFMs).
  • the computer program instructions when executed on or more processors implement the method including, generating benign labeled supplemental training example sequence pairs that include a start location, through a target amino acid location, to an end location. Each supplemental sequence pair matches the start location and the end location of a missense training example sequence pair.
  • the method includes inputting a supplemental training PFM with each supplemental sequence pair that is identical to the PFM of the missense training example sequence pair at the matching start and end location.
  • the method includes training the neural network-implemented model using the benign training example sequence pairs and the supplemental training example PFMs, and the missense training example sequence pairs, and the PFMs of the missense training at the matching start and end locations s.
  • the training influence of the training PFMs is attenuated during the training.
  • Computer readable media (CRM) implementations of the technology disclosed include one or more a non-transitory computer readable storage media impressed with computer program instructions, when executed on one or more processors, implement the method described above.
  • This CRM implementation includes one or more of the following features.
  • CRM implementation can also include features described in connection with system and method disclosed above.
  • FIG. 15 is a simplified block diagram 1500 of a computer system that can be used to implement the technology disclosed.
  • Computer system typically includes at least one processor that communicates with a number of peripheral devices via bus subsystem.
  • peripheral devices can include a storage subsystem including, for example, memory devices and a file storage subsystem, user interface input devices, user interface output devices, and a network interface subsystem.
  • the input and output devices allow user interaction with computer system.
  • Network interface subsystem provides an interface to outside networks, including an interface to corresponding interface devices in other computer systems.
  • the neural network such as variant pathogenicity classifier 157, the PFM calculator 184 and the input encoder 186 are communicably linked to the storage subsystem and user interface input devices.
  • User interface input devices can include a keyboard; pointing devices such as a mouse, trackball, touchpad, or graphics tablet; a scanner; a touch screen incorporated into the display; audio input devices such as voice recognition systems and microphones; and other types of input devices.
  • use of the term“input device” is intended to include all possible types of devices and ways to input information into computer system.
  • User interface output devices can include a display subsystem, a printer, a fax machine, or non-visual displays such as audio output devices.
  • the display subsystem can include a cathode ray tube (CRT), a flat-panel device such as a liquid crystal display (LCD), a projection device, or some other mechanism for creating a visible image.
  • the display subsystem can also provide a non-visual display such as audio output devices.
  • output device is intended to include all possible types of devices and ways to output information from computer system to the user or to another machine or computer system.
  • Storage subsystem stores programming and data constructs that provide the functionality of some or all of the modules and methods described herein. These software modules are generally executed by processor alone or in combination with other processors.
  • Memory used in the storage subsystem can include a number of memories including a main random access memory (RAM) for storage of instructions and data during program execution and a read only memory (ROM) in which fixed instructions are stored.
  • a file storage subsystem can provide persistent storage for program and data files, and can include a hard disk drive, a floppy disk drive along with associated removable media, a CD- ROM drive, an optical drive, or removable media cartridges.
  • the modules implementing the functionality of certain implementations can be stored by file storage subsystem in the storage subsystem, or in other machines accessible by the processor.
  • Bus subsystem provides a mechanism for letting the various components and subsystems of computer system communicate with each other as intended. Although bus subsystem is shown schematically as a single bus, alternative implementations of the bus subsystem can use multiple busses.
  • Computer system itself can be of varying types including a personal computer, a portable computer, a workstation, a computer terminal, a network computer, a television, a mainframe, a server farm, a widely -distributed set of loosely networked computers, or any other data processing system or user device. Due to the ever-changing nature of computers and networks, the description of computer system depicted in FIG. 15 is intended only as a specific example for purposes of illustrating the technology disclosed. Many other configurations of computer system are possible having more or less components than the computer system depicted in FIG. 15.
  • the deep learning processors can be GPUs or FPGAs and can be hosted by deep learning cloud platforms such as Google Cloud Platform, Xilinx, and Cirrascale.
  • deep learning processors include Google’s Tensor Processing Unit (TPU), rackmount solutions like GX4 Rackmount Series, GX8 Rackmount Series, NVIDIA DGX-1, Microsoft’ Stratix V FPGA, Graphcore’s Intelligent Processor Unit (IPU), Qualcomm’s Zeroth platform with Snapdragon processors, NVIDIA’s Volta, NVIDIA’s DRIVE PX, NVIDIA’s JETSON TX1/TX2 MODULE, Intel’s Nirvana, Movidius VPU, Fujitsu DPI, ARM’s DynamicIQ, IBM TrueNorth, and others.

Abstract

La présente invention concerne des systèmes et des procédés permettant de réduire le surapprentissage de modèles mis en œuvre par réseau neuronal qui traitent des séquences d'acides aminés et des matrices de fréquences de position associées. Le système génère des paires de séquences supplémentaires d'exemples d'entraînement, marquées bénignes, qui comprennent un emplacement de départ, par un emplacement d'acide aminé cible, à un emplacement de fin. Une paire de séquences supplémentaires complète une paire de séquences d'exemples d'entraînement de faux-sens pathogènes ou bénins. Elle présente des acides aminés identiques dans une référence et une séquence alternée d'acides aminés. Le système comprend une logique permettant d'entrer, à l'aide de chaque paire de séquences supplémentaires, une matrice de fréquence de position (PFM) d'entraînement supplémentaire qui est identique à la PFM du faux-sens bénin ou pathogène à l'emplacement de départ et de fin correspondants. Le système comprend une logique permettant d'atténuer l'influence d'entraînement des PFM d'entraînement pendant l'entraînement du modèle mis en œuvre par réseau neuronal par inclusion des PFM d'exemples d'entraînement supplémentaires dans les données d'entraînement.
PCT/US2019/031621 2017-10-16 2019-05-09 Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond WO2020081122A1 (fr)

Priority Applications (12)

Application Number Priority Date Filing Date Title
CA3064223A CA3064223A1 (fr) 2018-10-15 2019-05-09 Techniques basees sur l`apprentissage profond d`apprentissage prealable de reseaux neuronaux a convolution profonde
AU2019272062A AU2019272062B2 (en) 2018-10-15 2019-05-09 Deep learning-based techniques for pre-training deep convolutional neural networks
SG11201911777QA SG11201911777QA (en) 2018-10-15 2019-05-09 Deep learning-based techniques for pre-training deep convolutional neural networks
JP2019567603A JP6888123B2 (ja) 2018-10-15 2019-05-09 深層畳み込みニューラルネットワークを事前訓練するための深層学習ベースの技術
EP23209885.5A EP4296899A3 (fr) 2017-10-16 2019-05-09 Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond
KR1020197038080A KR102165734B1 (ko) 2018-10-15 2019-05-09 심층 컨볼루션 신경망을 사전 훈련시키기 위한 심층 학습 기반 기술
NZ759665A NZ759665A (en) 2018-10-15 2019-05-09 Deep learning-based techniques for pre-training deep convolutional neural networks
CN201980003263.9A CN111328419B (zh) 2018-10-15 2019-05-09 基于神经网络实现的方法和系统
EP19729404.4A EP3659143B1 (fr) 2017-10-16 2019-05-09 Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond
IL271091A IL271091B (en) 2018-10-15 2019-12-02 Methods based on deep learning for pre-training deep and complex neural networks
IL282689A IL282689A (en) 2018-10-15 2021-04-27 A variant pathogenicity classifier that is trained to prevent overfitting of frequency matrices
AU2021269351A AU2021269351B2 (en) 2018-10-15 2021-11-17 Deep learning-based techniques for pre-training deep convolutional neural networks

Applications Claiming Priority (14)

Application Number Priority Date Filing Date Title
USPCT/US2018/055840 2018-10-15
PCT/US2018/055840 WO2019079166A1 (fr) 2017-10-16 2018-10-15 Techniques basées sur l'apprentissage profond d'apprentissage de réseaux neuronaux à convolution profonde
US16/160,986 US11315016B2 (en) 2017-10-16 2018-10-15 Deep convolutional neural networks for variant classification
USPCT/US2018/055878 2018-10-15
US16/160,968 US11798650B2 (en) 2017-10-16 2018-10-15 Semi-supervised learning for training an ensemble of deep convolutional neural networks
PCT/US2018/055878 WO2019079180A1 (fr) 2017-10-16 2018-10-15 Réseaux neuronaux à convolution profonde de classification de variants
US16/160,968 2018-10-15
US16/160,903 2018-10-15
PCT/US2018/055881 WO2019079182A1 (fr) 2017-10-16 2018-10-15 Apprentissage semi-supervisé pour l'apprentissage d'un ensemble de réseaux neuronaux à convolution profonde
USPCT/US2018/055881 2018-10-15
US16/160,986 2018-10-15
US16/160,903 US10423861B2 (en) 2017-10-16 2018-10-15 Deep learning-based techniques for training deep convolutional neural networks
US16/407,149 2019-05-08
US16/407,149 US10540591B2 (en) 2017-10-16 2019-05-08 Deep learning-based techniques for pre-training deep convolutional neural networks

Publications (1)

Publication Number Publication Date
WO2020081122A1 true WO2020081122A1 (fr) 2020-04-23

Family

ID=70283180

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2019/031621 WO2020081122A1 (fr) 2017-10-16 2019-05-09 Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l'apprentissage profond

Country Status (8)

Country Link
JP (3) JP6888123B2 (fr)
KR (1) KR102165734B1 (fr)
CN (2) CN111328419B (fr)
AU (2) AU2019272062B2 (fr)
IL (2) IL271091B (fr)
NZ (1) NZ759665A (fr)
SG (2) SG11201911777QA (fr)
WO (1) WO2020081122A1 (fr)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111680762A (zh) * 2018-11-27 2020-09-18 成都工业学院 中药材适生地的分类方法及装置
CN111830408A (zh) * 2020-06-23 2020-10-27 朗斯顿科技(北京)有限公司 一种基于边缘计算和深度学习的电机故障诊断系统及方法
CN113539354A (zh) * 2021-07-19 2021-10-22 浙江理工大学 一种高效预测革兰氏阴性菌ⅲ型和ⅳ型效应蛋白的方法
CN113836892A (zh) * 2021-09-08 2021-12-24 灵犀量子(北京)医疗科技有限公司 样本量数据提取方法、装置、电子设备及存储介质
CN117688785A (zh) * 2024-02-02 2024-03-12 东北大学 一种基于种植思想的全张量重力梯度数据反演方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102418073B1 (ko) * 2020-06-08 2022-07-06 고려대학교 산학협력단 인공지능 기반 비디오 투시 연하검사 자동화 분석 장치 및 방법
CN112003735B (zh) * 2020-07-28 2021-11-09 四川大学 一种感知风险的深度学习驱动的极限传输容量调整方法
CN112183088B (zh) * 2020-09-28 2023-11-21 云知声智能科技股份有限公司 词语层级确定的方法、模型构建方法、装置及设备
KR102279056B1 (ko) * 2021-01-19 2021-07-19 주식회사 쓰리빌리언 지식전이를 이용한 유전자변이의 병원성 예측 시스템
CN113822342B (zh) * 2021-09-02 2023-05-30 湖北工业大学 一种安全图卷积网络的文献分类方法及系统
CN113963746B (zh) * 2021-09-29 2023-09-19 西安交通大学 一种基于深度学习的基因组结构变异检测系统及方法
US20240087683A1 (en) * 2022-09-14 2024-03-14 Microsoft Technology Licensing, Llc Classification using a machine learning model trained with triplet loss
CN115662520B (zh) * 2022-10-27 2023-04-14 黑龙江金域医学检验实验室有限公司 Bcr/abl1融合基因的检测方法及相关设备
CN116153396A (zh) * 2023-04-21 2023-05-23 鲁东大学 一种基于迁移学习的非编码变异预测方法

Citations (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1991006678A1 (fr) 1989-10-26 1991-05-16 Sri International Sequençage d'adn
US5641658A (en) 1994-08-03 1997-06-24 Mosaic Technologies, Inc. Method for performing amplification of nucleic acid with two primers bound to a single solid support
US20020055100A1 (en) 1997-04-01 2002-05-09 Kawashima Eric H. Method of nucleic acid sequencing
US20040002090A1 (en) 2002-03-05 2004-01-01 Pascal Mayer Methods for detecting genome-wide sequence variations associated with a phenotype
WO2004018497A2 (fr) 2002-08-23 2004-03-04 Solexa Limited Nucleotides modifies
US20040096853A1 (en) 2000-12-08 2004-05-20 Pascal Mayer Isothermal amplification of nucleic acids on a solid support
US7057026B2 (en) 2001-12-04 2006-06-06 Solexa Limited Labelled nucleotides
US7115400B1 (en) 1998-09-30 2006-10-03 Solexa Ltd. Methods of nucleic acid amplification and sequencing
WO2007010252A1 (fr) 2005-07-20 2007-01-25 Solexa Limited Procede de sequencage d'une matrice de polynucleotide
US7211414B2 (en) 2000-12-01 2007-05-01 Visigen Biotechnologies, Inc. Enzymatic nucleic acid synthesis: compositions and methods for altering monomer incorporation fidelity
US20070099208A1 (en) 2005-06-15 2007-05-03 Radoje Drmanac Single molecule arrays for genetic and chemical analysis
US20070128624A1 (en) 2005-11-01 2007-06-07 Gormley Niall A Method of preparing libraries of template polynucleotides
WO2007123744A2 (fr) 2006-03-31 2007-11-01 Solexa, Inc. Systèmes et procédés pour analyse de séquençage par synthèse
US7315019B2 (en) 2004-09-17 2008-01-01 Pacific Biosciences Of California, Inc. Arrays of optical confinements and uses thereof
US20080009420A1 (en) 2006-03-17 2008-01-10 Schroth Gary P Isothermal methods for creating clonal single molecule arrays
US7329492B2 (en) 2000-07-07 2008-02-12 Visigen Biotechnologies, Inc. Methods for real-time single molecule sequence determination
US20080108082A1 (en) 2006-10-23 2008-05-08 Pacific Biosciences Of California, Inc. Polymerase enzymes and reagents for enhanced nucleic acid sequencing
US7405281B2 (en) 2005-09-29 2008-07-29 Pacific Biosciences Of California, Inc. Fluorescent nucleotide analogs and uses therefor
US20080280773A1 (en) 2004-12-13 2008-11-13 Milan Fedurco Method of Nucleotide Detection
US20090088327A1 (en) 2006-10-06 2009-04-02 Roberto Rigatti Method for sequencing a polynucleotide template
US20130296175A1 (en) 2011-01-13 2013-11-07 Illumina Inc. Genetic Variants as Markers for Use in Urinary Bladder Cancer Risk Assessment, Diagnosis, Prognosis and Treatment
US20160085910A1 (en) 2014-09-18 2016-03-24 Illumina, Inc. Methods and systems for analyzing nucleic acid sequencing data
US20160357903A1 (en) * 2013-09-20 2016-12-08 University Of Washington Through Its Center For Commercialization A framework for determining the relative effect of genetic variants

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB0006153D0 (en) * 2000-03-14 2000-05-03 Inpharmatica Ltd Database
MX2011000451A (es) * 2008-07-16 2011-10-12 Dana Farber Cancer Inst Inc Distintivos y determinantes asociados con cancer de próstata y métodos para utilizarlos.
CN102171699B (zh) * 2008-10-02 2015-01-14 皇家飞利浦电子股份有限公司 确定用于从临床数据获得的标识的可靠性指标的方法和该可靠性指标用于支持一个标识胜过其它标识的用途
JP5773406B2 (ja) 2010-07-28 2015-09-02 学校法人明治大学 Ｇｐｉアンカー型タンパク質の判定装置、判定方法及び判定プログラム
MX2017014560A (es) * 2015-05-15 2018-03-01 Pioneer Hi Bred Int Caracterizacion rapida de sistemas de endonucleasa cas, secuencias pam y elementos arn de guia.
AU2016284455A1 (en) * 2015-06-22 2017-11-23 Myriad Women's Health, Inc. Methods of predicting pathogenicity of genetic sequence variants
CN107622182B (zh) * 2017-08-04 2020-10-09 中南大学 蛋白质局部结构特征的预测方法及系统
CN108197427B (zh) * 2018-01-02 2020-09-04 山东师范大学 基于深度卷积神经网络的蛋白质亚细胞定位方法和装置
CN108595909A (zh) * 2018-03-29 2018-09-28 山东师范大学 基于集成分类器的ta蛋白靶向预测方法

Patent Citations (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1991006678A1 (fr) 1989-10-26 1991-05-16 Sri International Sequençage d'adn
US5641658A (en) 1994-08-03 1997-06-24 Mosaic Technologies, Inc. Method for performing amplification of nucleic acid with two primers bound to a single solid support
US20020055100A1 (en) 1997-04-01 2002-05-09 Kawashima Eric H. Method of nucleic acid sequencing
US7115400B1 (en) 1998-09-30 2006-10-03 Solexa Ltd. Methods of nucleic acid amplification and sequencing
US7329492B2 (en) 2000-07-07 2008-02-12 Visigen Biotechnologies, Inc. Methods for real-time single molecule sequence determination
US7211414B2 (en) 2000-12-01 2007-05-01 Visigen Biotechnologies, Inc. Enzymatic nucleic acid synthesis: compositions and methods for altering monomer incorporation fidelity
US20040096853A1 (en) 2000-12-08 2004-05-20 Pascal Mayer Isothermal amplification of nucleic acids on a solid support
US7057026B2 (en) 2001-12-04 2006-06-06 Solexa Limited Labelled nucleotides
US20040002090A1 (en) 2002-03-05 2004-01-01 Pascal Mayer Methods for detecting genome-wide sequence variations associated with a phenotype
US20070166705A1 (en) 2002-08-23 2007-07-19 John Milton Modified nucleotides
WO2004018497A2 (fr) 2002-08-23 2004-03-04 Solexa Limited Nucleotides modifies
US7315019B2 (en) 2004-09-17 2008-01-01 Pacific Biosciences Of California, Inc. Arrays of optical confinements and uses thereof
US20080280773A1 (en) 2004-12-13 2008-11-13 Milan Fedurco Method of Nucleotide Detection
US20070099208A1 (en) 2005-06-15 2007-05-03 Radoje Drmanac Single molecule arrays for genetic and chemical analysis
WO2007010252A1 (fr) 2005-07-20 2007-01-25 Solexa Limited Procede de sequencage d'une matrice de polynucleotide
US7405281B2 (en) 2005-09-29 2008-07-29 Pacific Biosciences Of California, Inc. Fluorescent nucleotide analogs and uses therefor
US20070128624A1 (en) 2005-11-01 2007-06-07 Gormley Niall A Method of preparing libraries of template polynucleotides
US20080009420A1 (en) 2006-03-17 2008-01-10 Schroth Gary P Isothermal methods for creating clonal single molecule arrays
WO2007123744A2 (fr) 2006-03-31 2007-11-01 Solexa, Inc. Systèmes et procédés pour analyse de séquençage par synthèse
US20090088327A1 (en) 2006-10-06 2009-04-02 Roberto Rigatti Method for sequencing a polynucleotide template
US20080108082A1 (en) 2006-10-23 2008-05-08 Pacific Biosciences Of California, Inc. Polymerase enzymes and reagents for enhanced nucleic acid sequencing
US20130296175A1 (en) 2011-01-13 2013-11-07 Illumina Inc. Genetic Variants as Markers for Use in Urinary Bladder Cancer Risk Assessment, Diagnosis, Prognosis and Treatment
US20160357903A1 (en) * 2013-09-20 2016-12-08 University Of Washington Through Its Center For Commercialization A framework for determining the relative effect of genetic variants
US20160085910A1 (en) 2014-09-18 2016-03-24 Illumina, Inc. Methods and systems for analyzing nucleic acid sequencing data

Non-Patent Citations (21)

* Cited by examiner, † Cited by third party
Title
A. V. D. OORDS. DIELEMANH. ZENK. SIMONYANO. VINYALSA. GRAVESN. KALCHBRENNERA. SENIORK. KAVUKCUOGLU: "WAVENET: A GENERATIVE MODEL FOR RAW AUDIO", ARXIV:1609.03499, 2016
ANGERMUELLER CPARNAMAA TPARTS LSTEGLE O: "Deep Learning For Computational Biology", MOL SYST BIOL., vol. 12, 2016, pages 878, XP055540697, DOI: doi:10.15252/msb.20156651
BENTLEY ET AL., NATURE, vol. 456, 2008, pages 53 - 59
C. SZEGEDYW. LIUY. JIAP. SERMANETS. REEDD. ANGUELOVD. ERHANV. VANHOUCKEA. RABINOVICH: "GOING DEEPER WITH CONVOLUTIONS", ARXIV: 1409.4842, 2014
F. YUV. KOLTUN: "MULTI-SCALE CONTEXT AGGREGATION BY DILATED CONVOLUTIONS", ARXIV:1511.07122, 2016
G. HUANGZ. LIUL. VAN DER MAATENK. Q. WEINBERGER: "DENSELY CONNECTED CONVOLUTIONAL NETWORKS", ARXIV:1608.06993, 2017
H. CARTER ET AL: "Cancer-Specific High-Throughput Annotation of Somatic Mutations: Computational Prediction of Driver Missense Mutations", CANCER RESEARCH, vol. 69, no. 16, 4 August 2009 (2009-08-04), US, pages 6660 - 6667, XP055609626, ISSN: 0008-5472, DOI: 10.1158/0008-5472.CAN-09-1133 *
I. J. GOODFELLOWD. WARDE-FARLEYM. MIRZAA. COURVILLEY. BENGIO: "AUTOREGRESSIVE MODEL BASED ON A DEEP CONVOLUTIONAL NEURAL NETWORK FOR AUDIO GENERATION", 2016, TAMPERE UNIVERSITY OF TECHNOLOGY, article "CONVOLUTIONAL NETWORKS"
J. GUZ. WANGJ. KUENL. MAA. SHAHROUDYB. SHUAIT. LIUX. WANGG. WANG: "RECENT ADVANCES IN CONVOLUTIONAL NEURAL NETWORKS", ARXIV: 1512.07108, 2017
J. M. WOLTERINKT. LEINERM. A. VIERGEVERI. ISGUM: "DILATED CONVOLUTIONAL NEURAL NETWORKS FOR CARDIOVASCULAR MR SEGMENTATION IN CONGENITAL HEART DISEASE", ARXIV: 1704.03669, 2017
K. HEX. ZHANGS. RENJ. SUN: "DEEP RESIDUAL LEARNING FOR IMAGE RECOGNITION", ARXIV: 1512.03385, 2015
LEUNG MKDELONG AALIPANAHI B ET AL., MACHINE LEARNING IN GENOMIC MEDICINE: A REVIEW OF COMPUTATIONAL PROBLEMS AND DATA SETS, 2016
LIBBRECHT MWNOBLE WS: "Machine Learning Applications In Genetics and Genomics", NATURE REVIEWS GENETICS, vol. 16, no. 6, 2015, pages 321 - 32
LIZARDI ET AL., NAT. GENET., vol. 19, 1998, pages 225 - 232
MIN, S.LEE, B.YOON, S.: "Deep Learning In Bioinformatics", BRIEF. BIOINFORM. BBW068, 2016
PARK YKELLIS M: "Deep Learning For Regulatory Genomics", NAT. BIOTECHNOL., vol. 33, 2015, pages 825 - 826
R.K. SRIVASTAVAK. GREFFJ. SCHMIDHUBER: "HIGHWAY NETWORKS", ARXIV: 1505.00387, 2015
S. IOFFEC. SZEGEDY: "BATCH NORMALIZATION: ACCELERATING DEEP NETWORK TRAINING BY REDUCING INTERNAL COVARIATE SHIFT", ARXIV: 1502.03167, 2015
S. O. ARIKM. CHRZANOWSKIA. COATESG. DIAMOSA. GIBIANSKYY. KANGX. LIJ. MILLERA. NGJ. RAIMAN: "DEEP VOICE: REAL-TIME NEURAL TEXT-TO-SPEECH", ARXIV: 1702.07825, 2017
SUNDARAM LAKSSHMAN ET AL: "Predicting the clinical impact of human mutation with deep neural networks", NATURE GENETICS, NATURE PUBLISHING GROUP, NEW YORK, US, vol. 50, no. 8, 23 July 2018 (2018-07-23), pages 1161 - 1170, XP036559548, ISSN: 1061-4036, [retrieved on 20180723], DOI: 10.1038/S41588-018-0167-Z *
T. CHING ET AL., OPPORTUNITIES AND OBSTACLES FOR DEEP LEARNING IN BIOLOGY AND MEDICINE, 2017, Retrieved from the Internet <URL:www.biorxiv.org:142760>

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111680762A (zh) * 2018-11-27 2020-09-18 成都工业学院 中药材适生地的分类方法及装置
CN111680762B (zh) * 2018-11-27 2023-08-04 成都大学 中药材适生地的分类方法及装置
CN111830408A (zh) * 2020-06-23 2020-10-27 朗斯顿科技(北京)有限公司 一种基于边缘计算和深度学习的电机故障诊断系统及方法
CN111830408B (zh) * 2020-06-23 2023-04-18 朗斯顿科技(北京)有限公司 一种基于边缘计算和深度学习的电机故障诊断系统及方法
CN113539354A (zh) * 2021-07-19 2021-10-22 浙江理工大学 一种高效预测革兰氏阴性菌ⅲ型和ⅳ型效应蛋白的方法
CN113539354B (zh) * 2021-07-19 2023-10-27 浙江理工大学 一种高效预测革兰氏阴性菌ⅲ型和ⅳ型效应蛋白的方法
CN113836892A (zh) * 2021-09-08 2021-12-24 灵犀量子(北京)医疗科技有限公司 样本量数据提取方法、装置、电子设备及存储介质
CN113836892B (zh) * 2021-09-08 2023-08-08 灵犀量子(北京)医疗科技有限公司 样本量数据提取方法、装置、电子设备及存储介质
CN117688785A (zh) * 2024-02-02 2024-03-12 东北大学 一种基于种植思想的全张量重力梯度数据反演方法
CN117688785B (zh) * 2024-02-02 2024-04-16 东北大学 一种基于种植思想的全张量重力梯度数据反演方法

Also Published As

Publication number Publication date
AU2021269351A1 (en) 2021-12-09
KR102165734B1 (ko) 2020-10-14
KR20200044731A (ko) 2020-04-29
AU2021269351B2 (en) 2023-12-14
SG10202108013QA (en) 2021-09-29
NZ759665A (en) 2022-07-01
JP2021501923A (ja) 2021-01-21
IL271091B (en) 2021-05-31
CN113705585A (zh) 2021-11-26
JP7200294B2 (ja) 2023-01-06
JP2021152907A (ja) 2021-09-30
AU2019272062A1 (en) 2020-04-30
IL271091A (en) 2020-04-30
IL282689A (en) 2021-06-30
AU2019272062B2 (en) 2021-08-19
CN111328419A (zh) 2020-06-23
SG11201911777QA (en) 2020-05-28
CN111328419B (zh) 2021-10-19
JP6888123B2 (ja) 2021-06-16
JP2023052011A (ja) 2023-04-11

Similar Documents

Publication Publication Date Title
EP3659143B1 (fr) Techniques de pré-entraînement de réseaux neuronaux à convolution profonde fondées sur l&#39;apprentissage profond
US10540591B2 (en) Deep learning-based techniques for pre-training deep convolutional neural networks
AU2019272062B2 (en) Deep learning-based techniques for pre-training deep convolutional neural networks
CA3064223A1 (fr) Techniques basees sur l`apprentissage profond d`apprentissage prealable de reseaux neuronaux a convolution profonde
NZ788839A (en) Deep learning-based techniques for pre-training deep convolutional neural networks
NZ788045A (en) Deep convolutional neural networks for variant classification

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2019567603

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2019729404

Country of ref document: EP

Effective date: 20191209

ENP Entry into the national phase

Ref document number: 2019272062

Country of ref document: AU

Date of ref document: 20190509

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19729404

Country of ref document: EP

Kind code of ref document: A1

DPE1 Request for preliminary examination filed after expiration of 19th month from priority date (pct application filed from 20040101)
NENP Non-entry into the national phase

Ref country code: DE