WO2018094360A2 - Methods and systems for predicting dna accessibility in the pan-cancer genome - Google Patents
Methods and systems for predicting dna accessibility in the pan-cancer genome Download PDFInfo
- Publication number
- WO2018094360A2 WO2018094360A2 PCT/US2017/062626 US2017062626W WO2018094360A2 WO 2018094360 A2 WO2018094360 A2 WO 2018094360A2 US 2017062626 W US2017062626 W US 2017062626W WO 2018094360 A2 WO2018094360 A2 WO 2018094360A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- neural network
- data files
- rna
- seq
- data
- Prior art date
Links
Classifications
-
- C—CHEMISTRY; METALLURGY
- C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
- C12Q—MEASURING OR TESTING PROCESSES INVOLVING ENZYMES, NUCLEIC ACIDS OR MICROORGANISMS; COMPOSITIONS OR TEST PAPERS THEREFOR; PROCESSES OF PREPARING SUCH COMPOSITIONS; CONDITION-RESPONSIVE CONTROL IN MICROBIOLOGICAL OR ENZYMOLOGICAL PROCESSES
- C12Q1/00—Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions therefor; Processes of preparing such compositions
- C12Q1/68—Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions therefor; Processes of preparing such compositions involving nucleic acids
-
- 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
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic 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
- G16B25/00—ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
- G16B25/10—Gene or protein expression profiling; Expression-ratio estimation or normalisation
-
- 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
-
- 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
Definitions
- This disclosure relates generally to predicting DNA accessibility in a genomic sample, and more specifically to using a neural network to predict DNA accessibility in a genomic sample.
- DNA accessibility plays a key role in the regulatory machinery of DNA transcriptional events that can promote tumor growth.
- Locations where DNA is not tightly bound in nucleosomes detectable as DNase I hypersensitivity (DHS) sites, can render a DNA sequence accessible to other DNA-binding proteins, including a wide range of transcription factors (TFs).
- DHS sites are cell specific and play a crucial role in determining cell-selective transcriptional events.
- GWAS genome wide association studies
- RNA-seq data as a signal for cell type clustering and classification.
- a neural network Given paired RNA-seq and DNase-seq input data, a neural network is configured to learn to appropriately modulate its prediction to eliminate the need for a distinct trained model or unique output per cell type. As such, for the first time, accurate DNA accessibility predictions can be made for previously unseen cell types whose gene expressions are similar but unique from samples in the training data.
- genomic sample data including DNase-seq data files and
- RNA-seq data files for a plurality of cell types is obtained. Paired data files are generated from the genomic sample data by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype.
- a neural network is configured to be trained to predict DNA accessibility based on RNA-seq data using a plurality of batches of the paired data files, where configuring the neural network comprises configuring convolutional layers of the neural network to process a first input comprising DNA sequence data from one of the paired data files to generate a convolved output, and fully connected layers of the neural network following the convolutional layers to concatenate the convolved output with a second input comprising gene expression levels derived from RNA-seq data from the one of the paired data files and process the concatenation to generate a DNA accessibility prediction output.
- the first input may comprise a 600-base pair segment of DNA, and the gene expression levels may correspond to a selected subset of genes.
- the DNA accessibility prediction output may be a single prediction.
- the neural network is trained using the plurality of batches of the paired data files, and a computing device is configured to use the trained neural network to predict DNA accessibility in a genomic sample input comprising RNA-seq data and whole genome sequencing for a new cell type with respect to the genomic sample data.
- the genomic sample input may be associated with a cancer cohort from The Cancer Genome Atlas (TCGA) or a tumor.
- TCGA Cancer Genome Atlas
- the genomic sample data may be obtained from at least one of ENCODE project data and Roadmap Epigenomics project data.
- the RNA-seq data files may include data files having one or more of RNA-seq, poly A mRNA, poly A depleted, and single cell ENCODE labels, and RNA-seq data files that include error audit flags from the genomic sample data may be removed.
- the paired data files may be generated by assigning DNase-seq data files to RNA-seq data files based on matching biosample accessions or being from at least one of a same tissue sample, same cell line, or same patient.
- the paired data files may also be generated by randomly assigning a DNase-seq data file to one of a plurality of RNA-seq data files determined to be within a same biotype.
- the neural network may comprise a hierarchical structure of a plurality of convolutional layers each succeeded by a max-pooling layer, and the
- hierarchical structure may comprise at least three convolutional layers.
- the neural network may further comprise at least two fully connected layers following the hierarchical structure.
- training the neural network may comprise increasing a dynamic decay rate over a course of training when moving averages are updated for batch normalization, and using an adaptive moment estimation (Adam) optimization algorithm to optimize one or more network parameters of the neural network.
- Adam adaptive moment estimation
- the neural network may comprise a deep convolutional neural network, or a densely connected convolutional neural network.
- a convolutional neural network system comprises a sequence of neural network layers comprising a hierarchical structure of a plurality of convolutional layers each succeeded by a max-pooling layer.
- the hierarchical structure is configured to receive a first input comprising DNA sequence data from a paired data file and process the first input to generate a convolved output.
- the paired data file is generated from genomic sample data for a plurality of cell types by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype.
- the hierarchical structure may comprise at least three convolutional layers.
- At least two fully connected layers follow the hierarchical structure, and the at least two fully connected layers are configured to concatenate the convolved output with a second input comprising gene expression levels derived from RNA-seq data from the paired data file and process the concatenation to generate a DNA accessibility prediction output, that may be a single prediction.
- the sequence of neural network layers may be trained to predict DNA accessibility based on RNA-seq data using a plurality of batches of paired data files.
- a dynamic decay rate for the sequence of neural network layers may be configured to be increased over a course of training when moving averages are updated for batch normalization, and one or more network parameters of the sequence of neural network layers may be configured to be optimized based on an adaptive moment estimation (Adam) optimization algorithm.
- Adam adaptive moment estimation
- the sequence of neural network layers may comprise a deep convolutional neural network or a densely connected convolutional neural network.
- FIG. 1 illustrates an overview flow diagram of example operations for predicting
- FIG. 2 illustrates a block diagram of a system for predicting DNA accessibility using RNA-seq data in accordance with an embodiment.
- FIG. 3 illustrates a flow diagram of example operations for predicting DNA accessibility in a genomic sample in accordance with an embodiment.
- FIG. 4 illustrates a block diagram of a convolutional neural network system for predicting DNA accessibility in a genomic sample in accordance with an embodiment.
- FIG. 5 illustrates a flow diagram of a method of processing genomic sample data for a plurality of cell types using a convolutional neural network system in accordance with an embodiment.
- FIG. 6 illustrates a graphical representation of overall ROC AUC results for a validation dataset in accordance with an embodiment.
- FIG. 7 illustrates a graphical representation of overall ROC AUC results for a validation dataset after final dataset revision in accordance with an embodiment.
- FIG. 8 illustrates a graphical representation of PR AUC and ROC AUC results for a test dataset per whole genome sample in accordance with an embodiment.
- FIG. 9 illustrates a graphical representation of PR AUC results for a test dataset per whole genome sample in accordance with an embodiment.
- FIG. 10 illustrates a graphical representation of promoter and flank PR AUC
- FIG. 11 illustrates a graphical representation of overall PR AUC and ROC AUC results for a test dataset in accordance with an embodiment.
- FIG. 12 illustrates a graphical representation mutated promoter and flank sites normalized per number of patients analyzed per cohort in accordance with an embodiment.
- FIG. 13 illustrates a visual representation of box plots showing impact of mutations on predicted accessibility score at 600 base-pair promoter and flank sites in accordance with an embodiment.
- FIG. 14 illustrates a graphical representation of a fraction of mutated sites within a certain category of mutation that ended up flipped versus using the hgl9 reference genome in accordance with an embodiment.
- FIG. 15 illustrates a visual representation of DNA accessibility characteristics in accordance with an embodiment.
- FIG. 16 illustrates a block diagram of an exemplary client-server relationship that can be used for implementing one or more aspects of the various embodiments
- FIG. 17 illustrates a block diagram of a distributed computer system that can be used for implementing one or more aspects of the various embodiments.
- Coupled to is intended to include both direct coupling (in which two elements that are coupled to each other contact each other) and indirect coupling (in which at least one additional element is located between the two elements). Therefore, the terms “coupled to” and “coupled with” are used synonymously. Within the context of a networked environment where two or more components or devices are able to exchange data, the terms “coupled to” and “coupled with” are also used to mean “communicatively coupled with”, possibly via one or more intermediary devices.
- inventive subject matter is considered to include all possible combinations of the disclosed elements. As such, if one embodiment comprises elements A, B, and C, and another embodiment comprises elements B and D, then the inventive subject matter is also considered to include other remaining combinations of A, B, C, or D, even if not explicitly discussed herein.
- transitional term “comprising” means to have as parts or members, or to be those parts or members. As used herein, the transitional term “comprising” is inclusive or open-ended and does not exclude additional, unrecited elements or method steps.
- a server can include one or more computers operating as a web server, database server, or other type of computer server in a manner to fulfill described roles, responsibilities, or functions.
- programmed to is defined as one or more processors or cores of the computing element being programmed by a set of software instructions stored in the memory of the computing element to execute the set of functions on target data or data objects stored in the memory.
- any language directed to a computer should be read to include any suitable combination of computing devices, including servers, interfaces, systems, databases, agents, peers, engines, controllers, modules, or other types of computing devices operating individually or collectively.
- the computing devices comprise a processor configured to execute software instructions stored on a tangible, non-transitory computer readable storage medium (e.g., hard drive, FPGA, PLA, solid state drive, RAM, flash, ROM, etc.).
- the software instructions configure or program the computing device to provide the roles, responsibilities, or other functionality as discussed below with respect to the disclosed apparatus.
- the disclosed technologies can be embodied as a computer program product that includes a non-transitory computer readable medium storing the software instructions that causes a processor to execute the disclosed steps associated with implementations of computer- based algorithms, processes, methods, or other instructions.
- the various servers, systems, databases, or interfaces exchange data using standardized protocols or algorithms, possibly based on HTTP, HTTPS, AES, public-private key exchanges, web service APIs, known financial transaction protocols, or other electronic information exchanging methods.
- Data exchanges among devices can be conducted over a packet-switched network, the Internet, LAN, WAN, VPN, or other type of packet switched network; a circuit switched network; cell switched network; or other type of network.
- the focus of the disclosed inventive subject matter is to enable construction or configuration of a computing device to operate on vast quantities of digital data, beyond the capabilities of a human for purposes including predicting DNA accessibility in a genomic sample.
- each new type of genomic sample e.g., a biological cell or tissue for a given biotype
- DNase-seq DNase I hypersensitive site sequencing
- the Basset neural network model is one example of a cell-type specific model for predicting DNA accessibility.
- the Basset neural network model uses a binary matrix of genomic sample types and their respective DNA accessibilities as a universal list of potentially accessible genomic sites. Before training the Basset neural network model, the universal list is generated by agglomeratively clustering all overlapping DNase-seq peaks across all genomic samples.
- the final layer of the Basset neural network model is a multi-task output with a distinct prediction unit (output) for each biotype.
- a supplementary numerical signature that characterizes cells and tissues. Having such a cell signature as a parallel input can enable a neural network to leverage similarity and structure in the space of cell types and learn how DNA accessibility is modulated in a more general way (i.e., by a genomic sample's coordinates in the cell signature space).
- RNA-sequencing (RNA-seq) data i.e., the presence and quantity of RNA in a biological sample at a given moment in time, which is commonly available across large data sources of interest in research such as, for example, TCGA and the Genotype-Tissue Expression (GTEx) project.
- RNA-seq RNA-sequencing
- DNase-seq and microarray based gene expression levels from matched samples have been found to cluster similarly according to biological relationships, and many DNase I hypersensitivity (DHS) sites have been found to significantly correlate with gene expressions. Similar biologically meaningful neighborhood relationships also have appeared in both DNase-seq and RNA-seq data collected from the ENCODE project. Moreover, it has been observed that DNA accessibility is one of many complex factors that eventually determine gene expression at the level of RNA-seq, which makes the relationship between DNA accessibility and RNA-seq data not trivially invertible.
- DHS DNase I hypersensitivity
- FIG. 1 illustrates an overview flow diagram of example operations for predicting
- RNA-seq data DNA accessibility using RNA-seq data in accordance with an embodiment.
- a training dataset of genomic sample data comprising RNA-seq expression data files 102, DNA sequence data for all DNase I hypersensitivity (DHS) sites 104, and DNase-seq data files 106 for a plurality of cell types is used to train a neural network 108 to predict DNA accessibility based on RNA-seq data.
- neural network 108 is configured to process a first input comprising DNA sequence data and a second input comprising gene expression levels derived from RNA-seq data, where input DNase-seq and RNA-seq data files are paired based on a same biotype.
- a plurality of batches of paired DNase-seq and RNA-seq data files are used to train neural network 108.
- the neural network trained for predicting DNA accessibility 110 can be configured to receive RNA-seq data 112 and whole genome sequencing 114 for a new genomic sample input with respect to the training dataset, and predict DNA accessibility in the new genomic sample input 116.
- FIG. 2 illustrates a block diagram of a system for predicting DNA accessibility using RNA-seq data in accordance with an embodiment.
- elements for predicting DNA accessibility in a genomic sample include a training engine 210, a prediction engine 220, a persistent storage device 230, and a main memory device 240.
- training engine 210 may be configured to obtain genomic sample data related to a plurality of cell types, including RNA-seq expression data files 102, DNA sequence data for all DNase I hypersensitivity (DHS) sites 104, and DNase-seq data files 106, from either one or both of persistent storage device 230 and main memory device 240.
- DHS DNase I hypersensitivity
- Training engine 210 may then configure and train neural network 108, which may be stored in either one or both of persistent storage device 230 and main memory device 240, using the genomic sample data; and configure prediction engine 220 to use the trained neural network to predict DNA accessibility in a genomic sample input comprising RNA-seq data and whole genome sequencing for a new cell type with respect to the genomic sample data.
- prediction engine 220 may obtain RNA-seq data 112 and whole genome sequencing 114 for a new genomic sample input, and predict DNA accessibility in the genomic sample input 116 using the neural network trained for predicting DNA accessibility 110, which may be stored in either one or both of persistent storage device 230 and main memory device 240.
- any language directed to a training engine 210, a prediction engine 220, a persistent storage device 230 and a main memory device 240 should be read to include any suitable combination of computing devices, including servers, interfaces, systems, databases, agents, peers, engines, controllers, modules, or other types of computing devices operating individually or collectively to perform the functions ascribed to the various elements.
- computing devices including servers, interfaces, systems, databases, agents, peers, engines, controllers, modules, or other types of computing devices operating individually or collectively to perform the functions ascribed to the various elements.
- one or more of the functions of the system of FIG. 2 described herein may be performed within the context of a client-server relationship, such as by one or more servers, one or more client devices (e.g., one or more user devices) and/or by a combination of one or more servers and client devices.
- FIG. 3 further illustrates a flow diagram of example operations for predicting
- training engine 210 obtains genomic sample data including DNase-seq data files and RNA- seq data files for a plurality of cell types.
- RNA-seq are both available for a large and diverse collection of different cell types.
- the genomic sample data may be obtained from any human genomic data source, including from the
- RNA-seq data files selected from the ENCODE project database may include files having one or more of "RNA-seq”, “polyA mRNA”, “polyA depleted”, and
- RNA-seq data files that include ENCODE
- the genomic sample dataset is prepared for training a neural network to predict DNA accessibility based on RNA-seq data by generating a set of paired data files.
- the paired data files are generated from the genomic sample data by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype.
- the paired data files may be generated by assigning DNase-seq data files to RNA-seq data files based on matching biosample accessions.
- the paired data files also may be generated by randomly assigning a DNase-seq data file to one of a plurality of RNA-seq data files determined to be within a same biotype, e.g., in cases where a DNase-seq data file is determined to match several RNA-seq data files. In cases where multiple exact matches of biosample accession exist between the two file types, associations may be restricted to such exact matches. However, if exact match biosample accessions do not exist, RNA-seq and DNase-seq files may be associated based on being from, for example, at least one of a same tissue sample, same cell line, or same patient. Biotypes for which no such correspondences exist may be eliminated from the sample data.
- a neural network is configured to be trained to predict DNA accessibility based on RNA-seq data using a plurality of batches of the paired data files.
- the neural network for predicting DNA accessibility based on RNA-seq data includes a hierarchical structure comprising a plurality of convolutional layers each succeeded by a max-pooling layer.
- the neural network further includes at least two fully connected layers following the hierarchical structure.
- the neural network may comprise a deep convolutional neural network, or a densely connected convolutional neural network.
- configuring the neural network comprises configuring the convolutional layers to process a first input comprising DNA sequence data from one of the paired data files to generate a convolved output, and the fully connected layers following the convolutional layers to concatenate the convolved output with a second input comprising gene expression levels derived from RNA-seq data from the one of the paired data files and process the concatenation to generate a DNA accessibility prediction output.
- LINCS Library of Integrated Network-based Cellular Signatures
- the neural network is trained using the plurality of batches of the paired data files at step 306.
- data may be balanced per batch due to a selected ratio of negative training examples to positive training examples.
- Each batch may sample an equal amount of accessible and non-accessible sites without replacement, such that one pass through all available negative training examples constitutes multiple randomly permuted passes through all positive training examples.
- sites from the DNase-seq file may be randomly assigned to one of the plurality of corresponding RNA-seq expression vectors (derived gene expression levels) each time they are selected for a training batch.
- the batches of the paired data files may include a validation set for evaluating training progress. For example, a plurality of random samples may be selected from each of accessible and non-accessible sites per validation DNase-seq file and used to estimate an Area Under the Receiver Operating Characteristic curve (ROC AUC) throughout training. Prediction performance across whole genomes (i.e., all potential DHS sites) of all validation samples also may be evaluated. In cases where multiple RNA-seq file matches exist, predictions across the entire genome may be evaluated once for every possible DNase-seq and RNA-seq file pair, e.g., to characterize performance as captured by Precision Recall area under curve (PR AUC), which can be less misleading in the presence of data imbalance. Results on test sets may be evaluated across whole genomes following the same procedure.
- PR AUC Precision Recall area under curve
- the paired data files may comprise a plurality of unique biotypes and be partitioned into training, validation, and test sets as illustrated in Table 1.
- Table 1 Number of file types per dataset partition
- the validation set may be held constant, while the training and test sets may include a plurality of variations.
- the first test set may comprise randomly held-out samples, while the second test set may be selected such that all samples in the test set are from biotypes not represented in the training or validation data, e.g., to accurately simulate the application of the neural network described in the various embodiments herein.
- a greedy merge methodology may be used on all DNase-seq samples in the training sets to obtain a set of all potential sites of accessible DNA along the whole genome. For example, a fixed length, e.g., 600 base pairs centered around a DHS peak, may be used to define each site. Blacklisted sites, i.e., sites at which measurements have been deemed unreliable, may be excluded. The sequence for each genomic site may be obtained from a human genome database, e.g., the Genome Reference Consortium's human genome assembly hgl9.
- a dynamic decay rate for the sequence of neural network layers may be configured to be increased over a course of training when moving averages are updated for batch normalization, and one or more network parameters of the sequence of neural network layers may be configured to be optimized based on an adaptive moment estimation (Adam) optimization algorithm.
- Adam adaptive moment estimation
- a computing device e.g., prediction engine 220, is configured to use the trained neural network to predict DNA accessibility in a genomic sample input based on RNA-seq data for a new cell type with respect to the genomic sample (training) data.
- the genomic sample input may be associated with a cancer cohort from The Cancer Genome Atlas (TCGA) or a tumor.
- the cancer cohorts may include one or more of Lung Adenocarcinoma (LUAD), Lung Squamous Cell Carcinoma (LUSC), Kidney
- prediction engine 220 in operation may obtain a genomic sample input comprising RNA-seq data and whole genome sequencing for a new cell type with respect to the genomic sample data and, at step 310, predict DNA
- FIG. 4 illustrates a block diagram of a convolutional neural network system for predicting DNA accessibility in a genomic sample in accordance with an embodiment.
- Convolutional neural network system 400 includes a sequence of neural network layers comprising a hierarchical structure 402 comprising a plurality of convolutional layers each succeeded by a max-pooling layer.
- the hierarchical structure 402 is configured to receive a first input 404 comprising DNA sequence data from a paired data file and process the first input to generate a convolved output.
- first input 404 may be a 600 base-pair segment of DNA represented as a one-hot code (code having a single high ("1") bit and all other values low ("0")).
- the paired data file as described above, is generated from genomic sample data for a plurality of cell types by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype.
- the hierarchical structure 402 may comprise at least three
- convolutional layers (as shown), which apply a specified number of convolution filters to the data and, for each sub-region of the data, perform a set of mathematical operations to produce a single value in an output. Further, the first and second convolutional layers may be factorized to improve the rate of learning and final accuracy of system 400.
- At least two fully connected layers 406 follow the hierarchical structure 402 to perform a classification on the features extracted by the convolutional layers and down-sampled by the pooling layers.
- the at least two fully connected layers 406 are configured to concatenate the convolved output generated by the hierarchical structure 402 with a second input 408 comprising gene expression levels derived from RNA-seq data from the paired data file and process the concatenation to generate a single DNA accessibility prediction output 410.
- the sequence of neural network layers may be trained to predict DNA accessibility based on RNA-seq data using a plurality of batches of paired data files.
- batch normalization may be utilized at all layers, and a max norm constraint may be applied for regularization of all weights during the course of training.
- a dynamic decay rate may be used for the sequence of neural network layers for the purposes of achieving competitive performance more quickly than a fixed decay rate.
- the dynamic decay rate may be configured to increase over a course of training when moving averages are updated for batch normalization.
- an adaptive moment estimation (Adam) optimization algorithm, or one or more other optimization algorithms may be used to optimize one or more network parameters of the sequence of neural network layers.
- neural network system 400 is exemplary for implementing the embodiments herein, one skilled in the art will appreciate that various other neural network architectures (e.g., densely connected convolutional networks and Long Short- Term Memory Units (LSTMs)) and additions (such as attention mechanisms) may be utilized. As such, neural network system 400 should not be construed as being strictly limited to the embodiments described herein.
- LSTMs Long Short- Term Memory Units
- FIG. 5 illustrates a flow diagram of a method of processing genomic sample data for a plurality of cell types using the neural network system of FIG. 4.
- neural network system 400 may receive genomic sample data including DNase-seq data files and RNA- seq data files for plurality of cell types or, when trained, a genomic sample input comprising RNA-seq data and whole genome sequencing for a new cell type with respect to the genomic sample data.
- a first input comprising DNA sequence data from a paired data file is processed using a hierarchical structure comprising a plurality of convolutional layers (e.g., a layer which applies a specified number of convolution filters to the data and, for each sub-region of the data, performs a set of mathematical operations to produce a single value in an output) each succeeded by a max-pooling layer (e.g., a layer in which a down-sampling max filter is applied to sub-regions of the initial representation) to generate a convolved output.
- the paired data file is generated from the genomic sample by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype.
- At step 504 at least two fully connected layers (i.e., layers in which every node in the layer is connected to every node in the preceding layer) are configured to concatenate the convolved output with a second input comprising gene expression levels derived from RNA-seq data from the paired data file.
- the at least two fully connected layers process the concatenation to generate a single DNA accessibility prediction output.
- Several alternative versions of neural network system 400 were trained for testing purposes.
- cell-specific models were trained and evaluated following the procedure of the Basset neural network.
- DNase-seq peak data from 164 sample types obtained from the ENCODE and Roadmap Epigenomics projects was used for cell-specific model training, and a universal set of potential accessibility sites was created by a greedy merging of overlapping peaks across all DNase-seq data samples.
- a binary vector was used to label its accessibility state in each of the 164 cell types. The data was then split by genomic site so that 70,000 peak locations were held out for validation, 71,886 for testing, and the remaining 1.8 million sites were used for training.
- FIG. 6 illustrates overall ROC AUC for the small validation set over number of passes through all positive examples (positive epochs) for various model architectures using
- RNA-seq input RNA-seq input.
- Graph 600 illustrates the results from an experiment that added a fully connected (FC) layer of depth 500 before concatenating gene expressions with outputs from the convolutional layers.
- FC fully connected
- increasing the batch size and initializing the convolutional layers with weights from the final cell-specific model (transfer) improved performance most.
- Models trained on set 1 showed similar validation performance as those trained on set 2 with the same hyperparameters. This evaluation was done before the final dataset revision which revoked several suspected low-quality samples, yet still provided valuable feedback for model selection.
- FIG. 7 illustrates overall ROC AUC for the small validation set over positive training epochs for models trained after the final dataset revision.
- Graph 700 illustrates that a further increase in batch size as well as a decreased learning rate led to additional significant improvements.
- Changing the fraction of positive samples per training batch (from 0.5 to 0.25) also slightly improved both ROC AUC as well as PR AUC in whole genome validation.
- the transfer of weights learned before final revoking of data (FIG. 6) was a more effective initialization than transfer learning from the final cell-specific model. It was also confirmed that the same hyperparameters led to good validation performance across both training partitions.
- the cell-specific models had multi-task outputs so that each training sample provided an information rich gradient based on multiple labels for backpropagation.
- using RNA-seq inputs in neural network system 400 eliminated the need for multi-task outputs, so each sample only provided gradient feedback based on a single output.
- the batch size increase was thus intended to compensate for this change in output dimension to produce a more useful gradient for each batch.
- Neural network system 400 was initialized with weights learned from the prior iteration of the dataset, before the final revoked files were removed. In turn, those models were initialized with convolutional layer parameters from the best performing cell-specific model. An effective batch size of 2048 was used for training (two GPUs processing distinct batches of 1024), with an Adam learning rate of 0.0001 and a 0.25 fraction of positive to negative samples in every batch. [0080] Table 2 shows that final neural network system performance on the validation set, both overall and by biotype, was consistent across each of the two training partitions with respect to both ROC AUC as well as PR AUC.
- Table 2 Whole genome validation results for final neural network system trained on set 1 (tl) and set 2 (12)
- fibroblast of arm 0.898, 0.900 0.806, 0.809 0.898, 0.900 0.808, 0.811
- Table 3 and Table 4 summarize the results of applying neural network system 400 across whole genomes, at all potential DHS sites. For biotypes with more than a single file pair in the test set, each sample's results are listed.
- OCI-LY7 0.899, 0.899, 0.886, 0.886 0.654, 0.654, 0.655, 0.654
- hindlimb muscle 0.943 0.824
- Table 5 details the distribution of annotations applied to the 1.71 million sites considered in the held-out biotype training set, as well as the fraction of all positive samples that fall within each annotation type. Note that a single site may overlap with more than one annotation, and that Table 5 only reports details of the held-out biotypes partition (train/test set 2).
- graph 800 illustrates PR AUC and ROC AUC results for the test set of held-out biotypes (set 2) per whole genome sample. Since ROC AUC is affected by data imbalance, the PR AUC metric is a better evaluation of whole genome performance.
- graph 900 illustrates PR AUC results on the test set of held out biotypes
- FIGS. 10 and 1 1 confirm that the accuracy of these predictions was independent of whether the promoter and flank sites overlapped with the regions of genes used in our RNA- seq input vector.
- graph 1000 illustrates promoter and flank PR AUC results on the test set of held-out biotypes (set 2) split per whole genome sample and broken down by input gene set (L1000) membership. No clear performance difference was observed when promoter and flank regions were split into those that do and do not overlap the RNA-seq input gene set. Note that graph coloring is the same as defined in the legend of FIG. 8.
- graph 1 100 illustrates overall results on the test dataset of held out biotypes (set 2) broken down by site type and L1000 gene set membership.
- the neural network for predicting DNA accessibility described in the various embodiments herein can be applied to new datasets where RNA-seq 112 and whole genome sequence information 114 are available, as illustrated in FIG. 1.
- One application of the neural network system is to predict DNA accessibility for samples in the pan-cancer genome.
- To construct a predicted accessibility profile for each TCGA sample all somatic SNP, insertion (INS), and deletion (DEL) mutations were applied to any affected sites.
- INS somatic SNP
- DEL deletion
- graph 1200 illustrates the total number of SNP 1202, INDEL 1204, and SNP+INDEL 1206 mutations per cohort, normalized by each cohort's patient count.
- 3172 interest regions had a single SNP, 78 had 2 SNPs, and only 9 regions had between 3 and 5 SNPs.
- FIG. 13 illustrates a visual representation of box plots showing impact of mutations on predicted accessibility score at 600 base-pair promoter and flank sites in accordance with an embodiment.
- plot 1300 shows the distribution of changes due to SNPs only 1302,
- INDELs only 1304, and all mutations 1306 applied across all samples.
- INDEL mutations 1304 showed a larger variance in how much they impacted the accessibility score, which is to be expected since they typically impact a greater number of base pairs.
- FIG. 14 illustrates a graphical representation 1400 of a fraction of mutated sites within a certain category of mutation that ended up flipped versus using the hgl9 reference genome in accordance with an embodiment.
- graph 1400 it was investigated, applying the 80% precision threshold, how frequently each type of mutation caused accessibility decision changes.
- all mutations that led to changes in classification INS and DEL mutations were the most frequent causes of a decision flip.
- INDELs 1402 5.46% resulted in changed classification outcomes.
- FIG. 15 illustrates a visual representation 1500 of accessibility
- a neural network system as described herein was applied to six cancer cohorts: Lung Adenocarcinoma (LUAD), Lung Squamous Cell Carcinoma (LUSC), Kidney Chromophobe (KICH), Kidney Clear Cell
- RNA-seq data 1502 a Library of Integrated Cellular Signatures (LINCS) LI 000 gene expression platform gene expression level vectors from RNA-seq data 1502, raw predicted accessibility profile values 1502, and binarized accessibility profile data 1504 after applying the 80% precision threshold in samples from six TCGA cohorts.
- RNA-seq space 1506 a clear distinction can be seen between basal -like versus luminal A/B and HER2-enriched breast cancers (BRCA).
- BRCA basal -like versus luminal A/B and HER2-enriched breast cancers
- the lung (LUAD, LUSC) and breast (BRCA) cancer samples appear to have some common accessibility characteristics.
- the relationships between TCGA accessibility profiles visualized using t-SNE in FIG. 15 suggest that looking at cancers from the viewpoint of DNA accessibility offers different relationships and sub-categories than RNA-seq.
- ROC receiver operating characteristic
- PR precision-recall
- RNA-seq gene expression from RNA-seq can be added as a signature input that allows machine learning to exploit cell-type similarity.
- a neural network system for predicting DNA accessibility using RNA-seq data can achieve consistently high performance for predictions at promoter and flank regions of the genome, thus enabling a new tool for analysis of tumor genomes across different cell and tissue types and has provided the first glimpse of DNA accessibility (e.g., motor accessibility patterns) across several cohorts from The Cancer Genome Atlas (TCGA).
- TCGA Cancer Genome Atlas
- Systems, apparatus, and methods described herein may be implemented using digital circuitry, or using one or more computers using well-known computer processors, memory units, storage devices, computer software, and other components.
- a computer includes a processor for executing instructions and one or more memories for storing instructions and data.
- a computer may also include, or be coupled to, one or more mass storage devices, such as one or more magnetic disks, internal hard disks and removable disks, magneto- optical disks, optical disks, etc.
- Systems, apparatus, and methods described herein may be implemented using computers operating in a client-server relationship.
- the client computers are located remotely from the server computers and interact via a network.
- the client-server relationship may be defined and controlled by computer programs running on the respective client and server computers.
- Client-server relationship 1600 comprises client 1610 in communication with server 1620 via network 1630, and illustrates one possible division of DNA accessibility prediction tasks between client 1610 and server 1620.
- client 1610 in accordance with the various embodiments described above, may obtain genomic sample data including DNase-seq data files and RNA-seq data files for a plurality of cell types and send the genomic sample data to server 1620.
- Server 1620 may, in turn, receive genomic sample data/genomic sample input from client for DNA accessibility neural network training and prediction, generate paired data files from the genomic sample data by assigning DNase-seq data files to RNA-seq data files that are at least within a same biotype, configure a neural network to be trained to predict DNA accessibility based on RNA-seq data using a plurality of batches of the paired data files, and train the neural network to predict DNA accessibility based on RNA-seq data using a plurality of batches of the paired data files.
- Client 1610 may further send a genomic sample input comprising RNA-seq data and whole genome sequencing for a new cell type with respect to the genomic sample data to server 1620, which may receive the genomic sample input, predict DNA accessibility in the genomic sample input using the trained neural network, and send DNA accessibility prediction results for the genomic sample input to client 1610.
- server 1620 may receive the genomic sample input, predict DNA accessibility in the genomic sample input using the trained neural network, and send DNA accessibility prediction results for the genomic sample input to client 1610.
- client 1610 can include cellular smartphones, kiosks, personal data assistants, tablets, robots, vehicles, web cameras, or other types of computer devices.
- Systems, apparatus, and methods described herein may be implemented using a computer program product tangibly embodied in an information carrier, e.g., in a non-transitory machine-readable storage device, for execution by a programmable processor; and the method steps described herein, including one or more of the steps of FIGS. 3 and 5, may be implemented using one or more computer programs that are executable by such a processor.
- a computer program is a set of computer program instructions that can be used, directly or indirectly, in a computer to perform a certain activity or bring about a certain result.
- a computer program can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
- FIG. 17 A high-level block diagram of an exemplary apparatus that may be used to implement systems, apparatus and methods described herein is illustrated in Fig. 17. Apparatus
- 1700 comprises a processor 1710 operatively coupled to a persistent storage device 1720 and a main memory device 1730.
- Processor 1710 controls the overall operation of apparatus 1700 by executing computer program instructions that define such operations.
- the computer program instructions may be stored in persistent storage device 1720, or other computer-readable medium, and loaded into main memory device 1730 when execution of the computer program instructions is desired.
- training engine 210 and prediction engine 220 may comprise one or more components of computer 1700.
- the method steps of FIGS. 3 and 5 can be defined by the computer program instructions stored in main memory device 1730 and/or persistent storage device 1720 and controlled by processor 1710 executing the computer program instructions.
- the computer program instructions can be implemented as computer executable code programmed by one skilled in the art to perform an algorithm defined by the method steps of FIGS. 3 and 5. Accordingly, by executing the computer program instructions, the processor 1710 executes an algorithm defined by the method steps of FIGS. 3 and 5.
- Apparatus 1700 also includes one or more network interfaces 1780 for communicating with other devices via a network. Apparatus 1700 may also include one or more input/output devices 1790 that enable user interaction with apparatus 1700 (e.g., display, keyboard, mouse, speakers, buttons, etc.).
- apparatus 1700 e.g., display, keyboard, mouse, speakers, buttons, etc.
- Processor 1710 may include both general and special purpose microprocessors, and may be the sole processor or one of multiple processors of apparatus 1700.
- Processor 1710 may comprise one or more central processing units (CPUs), and one or more graphics processing units (GPUs), which, for example, may work separately from and/or multi-task with one or more CPUs to accelerate processing, e.g., for various deep learning and analytics applications described herein.
- processors 1710, persistent storage device 1720, and/or main memory device 1730 may include, be supplemented by, or incorporated in, one or more application-specific integrated circuits (ASICs) and/or one or more field programmable gate arrays (FPGAs).
- ASICs application-specific integrated circuits
- FPGAs field programmable gate arrays
- Persistent storage device 1720 and main memory device 1730 each comprise a tangible non-transitory computer readable storage medium.
- Persistent storage device 1720, and main memory device 1730 may each include high-speed random access memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), double data rate synchronous dynamic random access memory (DDR RAM), or other random access solid state memory devices, and may include non-volatile memory, such as one or more magnetic disk storage devices such as internal hard disks and removable disks, magneto-optical disk storage devices, optical disk storage devices, flash memory devices, semiconductor memory devices, such as erasable programmable read-only memory (EPROM), electrically erasable
- DRAM dynamic random access memory
- SRAM static random access memory
- DDR RAM double data rate synchronous dynamic random access memory
- EPROM erasable programmable read-only memory
- EEPROM programmable read-only memory
- CD-ROM compact disc read-only memory
- DVD-ROM digital versatile disc read-only memory
- Input/output devices 1790 may include peripherals, such as a printer, scanner, display screen, etc.
- input/output devices 1790 may include a display device such as a cathode ray tube (CRT), plasma or liquid crystal display (LCD) monitor for displaying information (e.g., a DNA accessibility prediction result) to a user, a keyboard, and a pointing device such as a mouse or a trackball by which the user can provide input to apparatus 1700.
- a display device such as a cathode ray tube (CRT), plasma or liquid crystal display (LCD) monitor for displaying information (e.g., a DNA accessibility prediction result) to a user, a keyboard, and a pointing device such as a mouse or a trackball by which the user can provide input to apparatus 1700.
- CTR cathode ray tube
- LCD liquid crystal display
- Any or all of the systems and apparatus discussed herein, including training engine 210 and prediction engine 220 may be performed by, and/or incorporated in, an apparatus such as apparatus 1700.
- FIG. 17 is a high-level representation of some of the components of such a computer for illustrative purposes.
- FIG. 17 is a high-level representation of some of the components of such a computer for illustrative purposes.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Biotechnology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Epidemiology (AREA)
- Bioethics (AREA)
- Public Health (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Organic Chemistry (AREA)
- Genetics & Genomics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Wood Science & Technology (AREA)
- Zoology (AREA)
- Proteomics, Peptides & Aminoacids (AREA)
- Analytical Chemistry (AREA)
- Microbiology (AREA)
- Immunology (AREA)
- Biochemistry (AREA)
- Pure & Applied Mathematics (AREA)
Abstract
Description
Claims
Priority Applications (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
AU2017362569A AU2017362569B2 (en) | 2016-11-18 | 2017-11-20 | Methods and systems for predicting DNA accessibility in the pan-cancer genome |
KR1020197016912A KR20190101966A (en) | 2016-11-18 | 2017-11-20 | Methods and Systems for Predicting DNA Accessibility in the Pan-Cancer Genome |
JP2019526482A JP2020501240A (en) | 2016-11-18 | 2017-11-20 | Methods and systems for predicting DNA accessibility in pan-cancer genomes |
CA3044254A CA3044254A1 (en) | 2016-11-18 | 2017-11-20 | Methods and systems for predicting dna accessibility in the pan-cancer genome |
EP17870742.8A EP3542296B1 (en) | 2016-11-18 | 2017-11-20 | Methods and systems for predicting dna accessibility in the pan-cancer genome |
IL266692A IL266692A (en) | 2016-11-18 | 2019-05-16 | Methods and systems for predicting dna accessibility in the pan-cancer genome |
Applications Claiming Priority (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US201662424370P | 2016-11-18 | 2016-11-18 | |
US62/424,370 | 2016-11-18 | ||
US201762481574P | 2017-04-04 | 2017-04-04 | |
US62/481,574 | 2017-04-04 | ||
US201762540523P | 2017-08-02 | 2017-08-02 | |
US62/540,523 | 2017-08-02 |
Publications (3)
Publication Number | Publication Date |
---|---|
WO2018094360A2 true WO2018094360A2 (en) | 2018-05-24 |
WO2018094360A3 WO2018094360A3 (en) | 2018-07-26 |
WO2018094360A9 WO2018094360A9 (en) | 2018-08-30 |
Family
ID=62145876
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2017/062626 WO2018094360A2 (en) | 2016-11-18 | 2017-11-20 | Methods and systems for predicting dna accessibility in the pan-cancer genome |
Country Status (8)
Country | Link |
---|---|
US (2) | US10467523B2 (en) |
EP (1) | EP3542296B1 (en) |
JP (1) | JP2020501240A (en) |
KR (1) | KR20190101966A (en) |
AU (1) | AU2017362569B2 (en) |
CA (1) | CA3044254A1 (en) |
IL (1) | IL266692A (en) |
WO (1) | WO2018094360A2 (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109271868A (en) * | 2018-08-21 | 2019-01-25 | 中国海洋大学 | Target recognition methods again based on intensive connection convolutional network suprasphere insertion |
CN110984062A (en) * | 2019-12-20 | 2020-04-10 | 华中科技大学 | Simulation scheduling method for large-scale reservoir group of watershed main and branch flows |
CN111916148A (en) * | 2020-08-13 | 2020-11-10 | 中国计量大学 | Method for predicting protein interaction |
JP2022521686A (en) * | 2019-02-11 | 2022-04-12 | フラッグシップ・パイオニアリング・イノベーションズ・ブイアイ,エルエルシー | Machine learning support polypeptide analysis |
US11443832B2 (en) | 2019-03-07 | 2022-09-13 | Nvidia Corporation | Genetic mutation detection using deep learning |
US11657897B2 (en) | 2018-12-31 | 2023-05-23 | Nvidia Corporation | Denoising ATAC-seq data with deep learning |
JP7492524B2 (en) | 2019-02-11 | 2024-05-29 | フラッグシップ・パイオニアリング・イノベーションズ・ブイアイ,エルエルシー | Machine learning assisted polypeptide analysis |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3542296B1 (en) | 2016-11-18 | 2021-04-14 | NantOmics, LLC | Methods and systems for predicting dna accessibility in the pan-cancer genome |
EP3801623A4 (en) * | 2018-06-01 | 2022-03-23 | Grail, LLC | Convolutional neural network systems and methods for data classification |
CN109063416B (en) * | 2018-07-23 | 2019-08-27 | 太原理工大学 | Gene expression prediction technique based on LSTM Recognition with Recurrent Neural Network |
CN108932567B (en) * | 2018-08-10 | 2020-12-01 | 燕山大学 | Convolutional neural network-based multi-energy-consumption index prediction method for cement sintering process |
CN113168886A (en) * | 2018-08-13 | 2021-07-23 | 豪夫迈·罗氏有限公司 | Systems and methods for germline and somatic variant calling using neural networks |
JP6471825B1 (en) | 2018-09-11 | 2019-02-20 | ソニー株式会社 | Information processing apparatus and information processing method |
CN109684742A (en) * | 2018-12-27 | 2019-04-26 | 上海理工大学 | A kind of frictional noise prediction technique based on BP neural network |
JP7159884B2 (en) * | 2019-01-23 | 2022-10-25 | ソニーグループ株式会社 | Information processing device and information processing method |
CN110277174B (en) * | 2019-06-14 | 2023-10-13 | 上海海洋大学 | Neural network-based prediction method for anticancer drug synergistic effect |
CN111048151B (en) * | 2019-11-19 | 2023-08-29 | 中国人民解放军疾病预防控制中心 | Virus subtype identification method and device, electronic equipment and storage medium |
TWI732370B (en) | 2019-12-04 | 2021-07-01 | 財團法人工業技術研究院 | Training device and training method for neural network model |
US11769332B2 (en) * | 2020-06-15 | 2023-09-26 | Lytx, Inc. | Sensor fusion for collision detection |
CN112530521B (en) * | 2020-12-03 | 2024-01-12 | 西安邮电大学 | SNP upper interaction identification method, system and application based on generation countermeasure network |
CN112951332A (en) * | 2021-02-25 | 2021-06-11 | 北京博富瑞基因诊断技术有限公司 | Method for severe intestinal tract aGVHD model based on aGVHD biomarker |
US11935627B2 (en) | 2021-12-29 | 2024-03-19 | Mujin, Inc. | System and method for text-based biological information processing with analysis refinement |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111394426B (en) * | 2013-05-23 | 2024-05-10 | 斯坦福大学托管董事会 | Transposition to natural chromatin for personal epigenomics |
CA3030453C (en) * | 2016-07-04 | 2024-01-02 | Deep Genomics Incorporated | Systems and methods for generating and training convolutional neural networks using biological sequences and relevance scores derived from structural, biochemical, population and evolutionary data |
EP3542296B1 (en) | 2016-11-18 | 2021-04-14 | NantOmics, LLC | Methods and systems for predicting dna accessibility in the pan-cancer genome |
AU2019206709B2 (en) * | 2018-01-15 | 2021-09-09 | Illumina Cambridge Limited | Deep learning-based variant classifier |
-
2017
- 2017-11-20 EP EP17870742.8A patent/EP3542296B1/en active Active
- 2017-11-20 AU AU2017362569A patent/AU2017362569B2/en not_active Expired - Fee Related
- 2017-11-20 KR KR1020197016912A patent/KR20190101966A/en not_active Application Discontinuation
- 2017-11-20 US US15/818,462 patent/US10467523B2/en active Active
- 2017-11-20 JP JP2019526482A patent/JP2020501240A/en not_active Ceased
- 2017-11-20 CA CA3044254A patent/CA3044254A1/en not_active Abandoned
- 2017-11-20 WO PCT/US2017/062626 patent/WO2018094360A2/en unknown
-
2019
- 2019-05-16 IL IL266692A patent/IL266692A/en unknown
- 2019-09-03 US US16/559,592 patent/US10748056B2/en active Active
Non-Patent Citations (4)
Title |
---|
ENCYCLOPEDIA OF DNA ELEMENTS (ENCODE |
KELLEY ET AL.: "asset: learning the regulatory code of the accessible genome with deep convolutional neural networks", GENOME RESEARCH, vol. 26, no. 7, 3 May 2016 (2016-05-03), pages 990 - 999, XP055507160, DOI: 10.1101/gr.200535.115 |
NATIONAL INSTITUTES OF HEALTH ROADMAP EPIGENOMICS MAPPING CONSORTIUM |
See also references of EP3542296A4 |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109271868A (en) * | 2018-08-21 | 2019-01-25 | 中国海洋大学 | Target recognition methods again based on intensive connection convolutional network suprasphere insertion |
US11657897B2 (en) | 2018-12-31 | 2023-05-23 | Nvidia Corporation | Denoising ATAC-seq data with deep learning |
JP2022521686A (en) * | 2019-02-11 | 2022-04-12 | フラッグシップ・パイオニアリング・イノベーションズ・ブイアイ,エルエルシー | Machine learning support polypeptide analysis |
JP7492524B2 (en) | 2019-02-11 | 2024-05-29 | フラッグシップ・パイオニアリング・イノベーションズ・ブイアイ,エルエルシー | Machine learning assisted polypeptide analysis |
US11443832B2 (en) | 2019-03-07 | 2022-09-13 | Nvidia Corporation | Genetic mutation detection using deep learning |
CN110984062A (en) * | 2019-12-20 | 2020-04-10 | 华中科技大学 | Simulation scheduling method for large-scale reservoir group of watershed main and branch flows |
CN111916148A (en) * | 2020-08-13 | 2020-11-10 | 中国计量大学 | Method for predicting protein interaction |
CN111916148B (en) * | 2020-08-13 | 2023-01-31 | 中国计量大学 | Method for predicting protein interaction |
Also Published As
Publication number | Publication date |
---|---|
US10467523B2 (en) | 2019-11-05 |
US20190392288A1 (en) | 2019-12-26 |
AU2017362569B2 (en) | 2020-08-06 |
US20180144261A1 (en) | 2018-05-24 |
IL266692A (en) | 2019-07-31 |
EP3542296A4 (en) | 2020-06-24 |
US10748056B2 (en) | 2020-08-18 |
EP3542296B1 (en) | 2021-04-14 |
JP2020501240A (en) | 2020-01-16 |
WO2018094360A3 (en) | 2018-07-26 |
WO2018094360A9 (en) | 2018-08-30 |
EP3542296A2 (en) | 2019-09-25 |
CA3044254A1 (en) | 2018-05-24 |
KR20190101966A (en) | 2019-09-02 |
AU2017362569A1 (en) | 2019-06-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
AU2017362569B2 (en) | Methods and systems for predicting DNA accessibility in the pan-cancer genome | |
Noble et al. | Spatial structure governs the mode of tumour evolution | |
Chen et al. | Random forests for genomic data analysis | |
Zakov et al. | An algorithmic approach for breakage-fusion-bridge detection in tumor genomes | |
McPherson et al. | ReMixT: clone-specific genomic structure estimation in cancer | |
AU2015101194A4 (en) | Semi-Supervised Learning Framework based on Cox and AFT Models with L1/2 Regularization for Patient’s Survival Prediction | |
US20220336049A1 (en) | Systems and methods for determining effects of therapies and genetic variation on polyadenylation site selection | |
Li et al. | A gene-based information gain method for detecting gene–gene interactions in case–control studies | |
Hosseini et al. | Smash++: an alignment-free and memory-efficient tool to find genomic rearrangements | |
Yin et al. | Ensemble variant interpretation methods to predict enzyme activity and assign pathogenicity in the CAGI4 NAGLU (Human N‐acetyl‐glucosaminidase) and UBE2I (Human SUMO‐ligase) challenges | |
Lock et al. | Bayesian genome-and epigenome-wide association studies with gene level dependence | |
Lee et al. | Survival prediction and variable selection with simultaneous shrinkage and grouping priors | |
Wnuk et al. | Deep learning implicitly handles tissue specific phenomena to predict tumor DNA accessibility and immune activity | |
Wnuk et al. | Predicting DNA accessibility in the pan-cancer tumor genome using RNA-seq, WGS, and deep learning | |
Dehiya et al. | Impact of structural prior knowledge in SNV prediction: Towards causal variant finding in rare disease | |
Mukherjee et al. | Predicting recurrence in clear cell Renal Cell Carcinoma: Analysis of TCGA data using outlier analysis and generalized matrix LVQ | |
Airey et al. | Identifying Genotype–Phenotype Correlations via Integrative Mutation Analysis | |
Huynh-Thu et al. | Gene regulatory network inference from systems genetics data using tree-based methods | |
CN114446393A (en) | Method, electronic device and computer storage medium for predicting liver cancer feature type | |
KR101810527B1 (en) | Algorithm for the construction of a regulatory network for more than 10,000 genes and method for the identification of causal genes in drug responses using the same algorithm | |
WO2021214774A1 (en) | Method and system for detecting mutational signatures and their exposures | |
Elmas et al. | The folded k-spectrum kernel: A machine learning approach to detecting transcription factor binding sites with gapped nucleotide dependencies | |
Zararsiz | Development and application of novel machine learning approaches for RNA-seq data classification | |
Wnuk et al. | Deep learning with implicit handling of tissue-specific phenomena predicts tumor DNA accessibility and immune activity | |
Chlis et al. | Extracting reliable gene expression signatures through stable bootstrap validation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17870742 Country of ref document: EP Kind code of ref document: A2 |
|
ENP | Entry into the national phase |
Ref document number: 3044254 Country of ref document: CA |
|
ENP | Entry into the national phase |
Ref document number: 2019526482 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 20197016912 Country of ref document: KR Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2017362569 Country of ref document: AU Date of ref document: 20171120 Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2017870742 Country of ref document: EP Effective date: 20190618 |