Hybrid variant calling

Abstract

A computer-implemented method for identifying a genomic variant is provided. The method includes obtaining one or more reference molecular sequences and sequencing data pertaining to a biological sample. One or more candidate variant positions are determined from the sequence reads for the biological sample. A classifier model is applied to each candidate variant position for selecting between a haplotype-aware or haplotype-agnostic variant analysis respectively. The classifier model is trained from haplotype structure and / or sequence reads identified with germline variants in a plurality of regions. Based on the application of the classifier model, respectively applying a haplotype-aware or haplotype-agnostic variant analysis to generate a variant identification for each candidate variant position.

Description

HYBRID VARIANT CALLINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This Application claims priority to United States Provisional Patent Application No. 63 / 685,009, filed on August 20, 2024, the contents of which is herein incorporated by reference in its entirety.BACKGROUND

[0002] Identifying genetic mutations using sequencing data is a crucial task in bioinformatics that has implications in diagnosis, prognosis, and treatment of multiple diseases including cancer. The task is a non-trivial one, especially when the sequencing technology used to generate the sequencing data has a high-error rate (which is common to single-molecule sequencing technologies) or when the mutations occur at low frequency (which is particularly true for cancer mutations) or in complex genomic regions. The low frequency of cancer mutations in a sequencing sample could be due to sample contamination (since tumor samples may contain some DNA from normal cells) or due to tumor heterogeneity.

[0003] Traditional approaches narrow down the scope of the problem by limiting to specific technologies and the type of mutations to detect. Therefore, different sets of tools and frameworks exist to identify germline and cancer mutations which are further stratified by whether the mutations are small or large. Additionally, the approaches used for low-error sequencing technologies (e.g., short-read sequencing) differ from the ones used for high-error sequencing technologies (e.g., single-molecule sequencing). The differences manifest primarily in data signatures as well as the statistical models used to differentiate errors from true mutations.

[0004] Advancing the use of explicit data signatures and statistical models can improve accuracy over traditional approaches as applied to varying sequencing technologies, but may come with unwieldly costs in computing power and time for general use. Accurate identification of germline variants, for example, may require significantly different analysis with different levels of time and computing power as compared to somatic variant identification and analysis. Furthermore, even the traditional approaches require parameter tuning to capture the specific characteristics of samples, which can depend on a multitude of factors including the technology used, sequencing protocol, sequencing sample, samplepurity and so on. Thus, there is a need in the art for improved systems, devices, and methods for identifying variants without overburdening computing resources.SUMMARY

[0005] The embodiments described herein relate to systems and methods for variant calling in bioinformatics workflows. More particularly, the embodiments described herein related toa bifurcated workflow that identifies genomic variants using either a haplotype- aware or a haplotype-agnostic algorithm.

[0006] In some embodiments, methods and systems are described that identify (or call) genomic variants, such as those associated with cancer and / or based on inherited traits, from patient molecular sequence data. Within sequencing data of a patient, candidate variant positions are determined from a plurality of sequence reads. In some embodiments, a classifier model is used to determine whether a haplotype-agnostic or haplotype-aware variant analysis (calling) approach is used to analyze the data to further identify properties of the candidate variant (e.g., signature, pathogenic, benign / cancerous). The classifier model is configured to indicate or recommend whether to use a haplotype- aware approach or haplotype-agnostic approach in further identifying properties of the candidate variant.

[0007] The classifier model may be trained with sequencing data associated with somatic variants, germline variants, and / or identification performance measures or gauges of haplotype-agnostic or haplotype-aware variant calling approaches (e.g., accuracy / speed), for example. Based on a given candidate variant position and / or other information about a candidate variant, the classifier outputs an indication of an analytical approach for further identifying the candidate variant (e.g., haplotype-aware or haplotype-agnostic). Based on the indication from the classifier model, a variant calling analytical approach is utilized to further analyze and predict properties of the candidate variant.

[0008] In some embodiments, a haplotype-agnostic approach is identified by the classifier model for further analyzing the candidate variant. A haplotype-agnostic approach may include obtaining normal sequence reads of a patient and tumor-sequence reads of a patient and converting the normal-sequence reads into a normal sample matrix and the tumorsequence reads into a tumor sample matrix. The matrices may be used as input (or for training) a convolutional neural network. The neural network may be trained with prior dataconfirmed to be associated with genetically normal or genetically varying tissue or bodily fluids (e.g., by cancer, inheritance, etc.).

[0009] In some embodiments, an augmentation process is utilized, in particular, for germline or haplotype- aware analysis. Errors and / or germline variants mimicking somatic variants including insertions can be common in some reading technologies. An analysis of a sequence may include obtaining reference molecular sequence(s) and augmenting them with space(s) inserted in the sequence to account for, and based on, common insertion occurrences from historical sequencing and alignment data around a window of a candidate variant and similarly augmenting the sequence data from the patient sample. In some embodiments, matrices representing a sample’s read sequence and reference sequence(s) are provided to a trained neural network and an output is obtained from the neural network predicting / identifying variant characteristics. In some embodiments, the classifier may be configured to indicate or recommend whether to use a haplotype- aware approach or haplotype-agnostic approach in further identifying properties of the candidate variant.

[0010] In accordance with a first aspect of the present disclosure, a computer- implemented method is provided for identifying a genomic variant. The method includes: obtaining one or more reference sequences; obtaining sequencing data comprising a plurality of sequence reads pertaining to a biological sample; determining one or more candidate variant positions from the plurality of sequence reads; for each candidate variant position, applying a classifier model for selecting use of a haplotype-aware or haplotype-agnostic variant analysis with respect to the candidate variant position; based on the output of the classifier model for each candidate variant position, respectively applying a haplotype-aware or haplotype-agnostic variant analysis for the corresponding candidate variant position; and transmitting an output variant identification for each candidate variant position based on the respectively applied haplotype-aware or haplotype-agnostic variant analysis.

[0011] In at least one embodiment of the first aspect, the classifier model is trained from sequencing data identified with somatic variants in a plurality of regions.

[0012] In at least one embodiment of the first aspect, the classifier model is trained using performance measures for identifying variants based on one or more candidate regions of interest.

[0013] In at least one embodiment of the first aspect, at least one of the haplotype-aware or haplotype-agnostic variant analysis comprises applying a neural network trained with sequencing data identified with variants in a plurality of regions.

[0014] In at least one embodiment of the first aspect, the haplotype-aware variant analysis includes: obtaining augmented sequence reads by inserting one or more spaces in the sequencing data; obtaining an augmented reference sequence by inserting one or more spaces in the one or more reference sequences; converting a segment of the augmented sequence reads around the respective candidate variant into a sample matrix; converting a segment of the augmented reference sequence around the respective candidate variant into a reference matrix; providing the sample matrix and the reference matrix to a trained neural network; and receiving, at the output of the trained neural network, prediction data relating to one or more variants identified within the sequencing data.

[0015] In at least one embodiment of the first aspect, augmenting the sequence reads and the reference sequence includes: detecting one or more inserted bases in the sequencing data; and, for each inserted base detected, inserting a space in the reference sequence at a position in the reference sequence corresponding to a position of the inserted base, and inserting a space at the position of the inserted base in every sequence read of the sequencing data for which no insertions were detected at the position of the inserted base.

[0016] In at least one embodiment of the first aspect, the trained neural network is provided with at least one of: a variant position matrix representing a respective candidate variant’s position within the segment of the augmented sequence reads; a coverage matrix representing coverage or depth of the segment of the augmented sequence reads; an alignment feature matrix representing an alignment feature of the augmented sequence reads; or a knowledgeable base matrix representing predetermined information about one or more variants.

[0017] In at least one embodiment of the first aspect, the sequencing data comprises: sequence reads from a normal sample of an individual and sequence reads from a tumor sample of the individual. The haplotype-agnostic variant analysis includes: converting a segment of the sequence reads for the normal sample and a segment of the sequence reads for the tumor sample into a normal sample matrix and a tumor sample matrix, respectively; and feeding the normal sample matrix and the tumor sample matrix into a trained convolutionalneural network. In an embodiment, the sequence reads for the tumor sample represent genetic information of tissue from a tumor biopsy.

[0018] In at least one embodiment of the first aspect, converting the segment of the sequence reads for the normal sample into the normal sample matrix comprises augmenting the segment of the sequence reads for the normal sample by inserting one or more spaces in one or more sequence reads for the normal sample. Similarly, converting the segment of the sequence reads for the tumor sample into the tumor sample matrix comprises augmenting the segment of the sequence reads for the tumor sample by inserting one or more spaces in one or more sequence reads for the tumor sample.

[0019] In accordance with a second aspect of the present disclosure, a system is provided for identifying a genomic variant in a biological sample. The system includes one or more processors configured to: obtain one or more reference molecular sequences; obtain sequence data comprising a plurality of sequence reads pertaining to the biological sample; determine one or more candidate variant positions from the plurality of sequence reads; for each candidate variant position, apply a classifier model for selecting use of a haplotype-aware or haplotype-agnostic variant analysis with respect to the candidate variant position; based on the output of the classifier model, respectively apply a haplotype-aware or haplotype-agnostic variant analysis for each candidate variant position; and transmit an output variant identification for each candidate variant position based on the respectively applied haplotype- aware or haplotype-agnostic variant analysis.

[0020] In at least one embodiment of the second aspect, the classifier model is trained from sequencing data identified with somatic variants in a plurality of regions.

[0021] In at least one embodiment of the second aspect, the classifier model is trained using performance measures for identifying variants based on one or more candidate regions of interest.

[0022] In at least one embodiment of the second aspect, at least one of the haplotype- aware or haplotype-agnostic variant analysis comprises applying a neural network trained with sequencing data identified with variants in a plurality of regions.

[0023] In at least one embodiment of the second aspect, the haplotype-aware variant analysis includes: obtaining augmented sequence reads by inserting one or more spaces in thesequencing data; obtaining an augmented reference sequence by inserting one or more spaces in the one or more reference sequences; converting a segment of the augmented sequence reads around the respective candidate variant into a sample matrix; converting a segment of the augmented reference sequence around the respective candidate variant into a reference matrix; providing the sample matrix and the reference matrix to a trained neural network; and receiving, at the output of the trained neural network, prediction data relating to one or more variants identified within the sequencing data.

[0024] In at least one embodiment of the second aspect, augmenting the sequence reads and the reference sequence includes: detecting one or more inserted bases in the sequencing data; and, for each inserted base detected, inserting a space in the reference sequence at a position in the reference sequence corresponding to a position of the inserted base and inserting a space at the position of the inserted base in every sequence read of the sequencing data for which no insertions were detected at the position of the inserted base.

[0025] In at least one embodiment of the second aspect, the trained neural network is provided with at least one of: a variant position matrix representing a respective candidate variant’s position within the segment of the augmented sequence reads; a coverage matrix representing coverage or depth of the segment of the augmented sequence reads; an alignment feature matrix representing an alignment feature of the augmented sequence reads; or a knowledgeable base matrix representing predetermined information about one or more variants.

[0026] In at least one embodiment of the second aspect, the sequencing data includes: sequence reads from a normal sample of an individual and sequence reads from a tumor sample of the individual. The haplotype-agnostic variant analysis includes: converting a segment of the sequence reads for the normal sample and a segment of the sequence reads for the tumor sample into a normal sample matrix and a tumor sample matrix, respectively; and feeding the normal sample matrix and the tumor sample matrix into a trained convolutional neural network.

[0027] In at least one embodiment of the second aspect, the sequence reads for the tumor sample represent genetic information of tissue from a tumor biopsy.

[0028] In at least one embodiment of the second aspect, converting the segment of the sequence reads for the normal sample into the normal sample matrix comprises augmenting the segment of the sequence reads for the normal sample by inserting one or more spaces in one or more sequence reads for the normal sample. Similarly, converting the segment of the sequence reads for the tumor sample into the tumor sample matrix comprises augmenting the segment of the sequence reads for the tumor sample by inserting one or more spaces in one or more sequence reads for the tumor sample.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The detailed description below is set forth with reference to the accompanying figures, which include the following.

[0030] FIG. 1 illustrates a process flow for calling a variant from a molecular sequence, in accordance with at least some embodiments.

[0031] FIG. 2 illustrates a process flow for programming / configuring a classifier model to select between a haplotype-aware or haplotype-agnostic variant calling workflow, in accordance with at least some embodiments.

[0032] FIG. 3 illustrates a method that can be performed as part of a haplotype-agnostic variant calling process, in accordance with at least some embodiments.

[0033] FIG. 4 illustrates a method that can be performed as part of a haplotype-aware variant calling process, in accordance with at least some embodiments.

[0034] FIG. 5 illustrates an exemplary architecture of a neural network for variant calling, in accordance with at least some embodiments.

[0035] FIG. 6 illustrates an example computer system that may be utilized to implement techniques disclosed herein.DETAILED DESCRIPTION

[0036] Disclosed herein are systems and methods including a classifier model that selects between haplotype-aware and haplotype-agnostic methods for calling a variant within sequencing data (e.g., a plurality of reads sequenced from a biological sample). The selectionperformed by the classifier model is based on a position and / or other information pertaining to a candidate variant identified within the sequencing data.

[0037] FIG. 1 illustrates a process flow (i.e., method) for calling a variant from molecular sequence data, in accordance with at least some embodiments. The method may be performed, at least in part, by one or more processors, such as but not limited to a central processing unit (CPU), graphics processing unit (GPU), tensor processing unit (TPU), application specific integrated circuit (ASIC), field programmable gate array (FPGA), or the like. The one or more processors may be communicatively coupled by a bus in a computer system and may optionally be connected to one or more computer-readable media such as volatile or non-volatile memory. An example of the computer system used to perform the method is set forth below and described with reference to FIG. 6. However, it will be recognized that any computer system capable of performing one or more steps of the method is within the scope of the present disclosure, including distributed computer systems in communication over a network or computer systems implemented within a sequencer instrument configured to generate the molecular sequence data.

[0038] At block 100, one or more reference molecular sequences are obtained. The reference molecular sequences may be obtained, for example, from one or more private or public repositories such as the Reference Sequence (RefSeq), an open-access, annotated and curated collection of nucleotide base sequences built by the National Center for Biotechnology Information (NCBI) or the NCBI Genomes FTP site storing a set of complete genomes of different organisms. It should be understood that in some embodiments, a specific copy of the reference sequence may be stored locally (e.g., in a system memory 72 of FIG. 6), while in other embodiments the reference sequence may be accessed through a remote server, e.g., through network 90. Furthermore, in some embodiments, the entire reference sequence may be obtained while in other embodiments one or more sections of the reference sequence may be obtained - e.g., only the section(s) that is / are associated with a particular assay. Thus, a “reference sequence” as used herein refers generally to one or more sections of the reference sequence, which may or may not include the entire reference sequence.

[0039] At block 110, a plurality of sequence reads corresponding to a genetic sample (e.g., a sample comprising a patient’s DNA or RNA material) as sequenced by a sequencing device 50 of FIG. 6 can be obtained. As discussed above, the sequence reads can be obtainedeither directly from sequencer instrument 90, or from one or more local or remote volatile or non-volatile memories, storage devices, or databases communicatively coupled to a computer system. The obtained sequence reads can be already pre-processed (e.g., pre-aligned) or they can be “raw,” in which case a method may also include a preprocessing (e.g., pre-aligning) step (not illustrated), which are well-known to those of skill in the art of bioinformatics. Also, while in some embodiments, entire sequence reads (as generated by sequencer instrument 90) can be obtained, in other embodiments only sections of the sequence reads are obtained. Thus, “obtaining a sequence read,” as used herein, refers generally to obtaining one or more sections of one or more (e.g., adjacent) sequence reads.

[0040] At block 120, one or more candidate variant positions within the sequence read can be obtained. In some embodiments, this includes scanning the plurality of obtained sequence reads, comparing them to the obtained reference sequence, and determining a set of one or more positions within the sequence read that appear to include variants of some type (e.g., insertion or deletion variants (INDELs), single-nucleotide variants (SNVs), or structural variants (SVs)). Finding the candidate variant position can include, for example, checking all positions in the reference and determining positions for which at least one of the sequence reads differ from the reference. In determining the candidate positions some filters can also be used, for example, filters on the number or the percentage of reads that must be different from the reference sequence for a particular position to be considered a candidate variant position. In other embodiments, the sequence reads could have already been scanned and analyzed (e.g., by a separate software and / or hardware module), in which case the plurality of candidate positions can be obtained from a memory or a database coupled to computer system 10.

[0041] At block 130, a classifier model analyzes each respective candidate variant position obtained at block 120. The classifier model uses information obtained about the candidate variant at each position and selects between a haplotype-aware or haplotypeagnostic method for further analyzing the candidate variant at that position. The classifier model may include a machine learning algorithm trained with other sequencing data in which variants of particular types were identified, failed to be accurately identified, the position of the variant, (variant calling) methods used in the identification process, and / or performance metrics pertaining to the variant calling methods (e.g., speed, accuracy, sensitivity). The types of variants used in the training process may be distributed between germline variants andsomatic variants, for example. In some embodiments, the variants are distributed between insertion / deletion variants (INDELs), single-nucleotide variants (SNVs), and / or structural variants (SVs). In some embodiments, the machine learning algorithm includes a neural network model such as a recurrent or convolutional neural network model, as further described herein.

[0042] Based on an input of sequence data collected from a sample, including information about a candidate variant, the classifier model selects a type of analysis for further characterizing the candidate variant. Information about the candidate variant may include the position of the candidate variant and other information such as likely type(s) (INDEL, SNV, SV). Analysis of such information about the candidate variant generates an indication of its likelihood to be a germline or somatic variant, for example, and the speed and / or accuracy of analysis options for further identifying the candidate variant. In some embodiments, the classifier model is programmed with thresholds for selecting or indicating optimal approaches based on predicted or predetermined speeds and accuracy of the approaches. In some embodiments, these thresholds can be adjusted within a user interface, such as based on the needs of a particular user.

[0043] At block 135, based on the classifier model selecting or indicating that a haplotype-aware analysis is optimal, a haplotype-aware analysis is performed on the candidate variant. At block 145, and as further described herein, a haplotype-aware analysis can include, at least in part, application of a deep learning neural network with multiple layers to account for the complexity in differentiating the candidate variant between potential types (e.g., germlines) and / or sequencing errors.

[0044] At block 140, based on the classifier model selecting or indicating that a haplotype-agnostic analysis is optimal, a haplotype-agnostic analysis is performed on the candidate variant. As further described herein, a haplotype-agnostic analysis may include, at least in part, application of a neural network less deeply trained that a haplotype-aware analysis and / or less convoluted operations between a reference sequence matrix and a sample matrix.

[0045] At block 150, based on the haplotype-agnostic analysis of the candidate variant, further predicted properties of the candidate variant are obtained. The information may include a variant type, structure, and / or other properties (e.g., cancer type, risk of diseasedevelopment). Predicted properties may include information accessed through external sources (e.g., variant registry) corresponding to the predicted variant type. The variant type may include information about whether the variant is germline or somatic. In some embodiments, the analysis includes a quality score or other quality measure associated with the predicted properties. If the quality measure crosses a particular threshold, the analysis may indicate a failure to further identify the candidate variant (e.g., indicates a likely error in the sequence data) and / or can trigger a further haplotype-aware analysis to be performed for the candidate variant (e.g., at block 135).

[0046] At block 135, based on the classifier model selecting or indicating that a haplotype-aware analysis of the candidate variant is optimal, a haplotype- aware analysis is performed. As further described herein, and in some embodiments, a haplotype-aware analysis performed at block 145 may include a deep-learning neural network trained with information pertaining to subtleties between different germline and somatic variants, and likely errors in sequencing data.

[0047] At block 155, based on the haplotype-aware analysis of the candidate variant, further predicted properties of the candidate variant are obtained. Properties and other information may be similar to that described with respect to the haplotype- agnostic analysis performed at block 140.

[0048] At block 160, the further identified properties / prediction(s) for the candidate variant (from blocks 150 and / or block 155) are reported. Reporting may include generating an electronic report and causing transmission / storage / display such as across a computer network to a remote computer or database storage system.

[0049] FIG. 2 illustrates a process flow (i.e., a method) for programming / configuring a classifier to select between a haplotype-aware or haplotype-agnostic variant calling analysis, in accordance with at least some embodiments. The method may be performed, at least in part, by one or more processors (similar to the method of FIG. 1 set forth above), which may be communicatively coupled by a bus in a computer system and may optionally be connected to one or more computer-readable media such as volatile or non-volatile memory. An example of the computer system used to perform the method is set forth below and described with reference to FIG. 6. However, it will be recognized that any computer system capable of performing one or more steps of the method is within the scope of the present disclosure,including distributed computer systems in communication over a network or computer systems implemented within a sequencer instrument configured to generate the molecular sequence data.

[0050] Classifier models include machine learning algorithms such as neural networks (e.g., CNNs, RNNs, etc.). At block 200, training of such a machine learning algorithm begins. At block 210, haplotype-specific data for training the classifier model is obtained, and may include haplotype structures and sequencing data with identified germline variants of different types in a plurality of regions. For example, the sequencing data may be labeled with particular germline variants previously identified.

[0051] At block 220, data not associated with germline variants (e.g., somatic variants or sequencing errors) is obtained for training the classifier model.

[0052] At block 230, a first model associated with a haplotype-aware approach is trained, configured, and / or otherwise obtained for identifying variants of multiple types (e.g., both germline and / or somatic). In some embodiments, the training data obtained at blocks 210 and 220 is used to train the first model in the haplotype-aware approach. In some embodiments, a method implemented for the haplotype-aware approach may include application of deeplearning model(s) particularly trained to differentiate between variants of particular structure and / or location and whether they are somatic, germline, or likely errors.

[0053] At block 240, a second model associated with a haplotype-agnostic approach is trained, configured, and / or otherwise obtained for identifying variants of multiple types (e.g., both germline and / or somatic). The second model may be configured to identify variants in regions or types that may be more readily identified without the need for the (deep) training required for differentiating between multiple possible variant types in a particular region and / or of a particular structure. In some embodiments, the haplotype-agnostic approach can thus accurately identify a variant without the same level of computing power needed for a haplotype-aware approach.

[0054] In some embodiments, the model(s) may be trained using sequencing data typical of that received from certain technology types (e.g., sanger, sequencing-by synthesis, nanopore, sequencing-by-expandomer, long- or short-sequence reads) to correspond with sequencing data used to identify variants from subsequent reads using similar technologytypes. The sequencing data may include (labeled) somatic or germline variants (or sequencing errors) of a variety of types and span across a plurality of regions of the genome.

[0055] At block 250, performance metrics of the different approaches (haplotype- aware and haplotype-agnostic) are determined for identification of different types, structures and / or positions of candidate variants. In some embodiments, performance metrics may include the accuracy, sensitivity, and / or speed of the approaches in identifying variants (or errors) based on candidate parameters / properties.

[0056] At block 260, a classifier model for selecting variant calling approaches is configured / programmed based on properties of candidate variants and performance metrics of the approaches. In some embodiments, the classifier model includes a machine learning model, such as a neural network, trained with the performance metrics and corresponding candidate variant properties. The classifier model is thereby trained to select between approaches based on properties of a candidate variant (e.g., structure, position) and the likely performance (e.g., speed, accuracy) of the respective approaches in further identifying a candidate variant.

[0057] FIG. 3 illustrates a method that can be performed as part of a haplotype-agnostic variant calling process, in accordance with at least some embodiments. In some embodiments, a neural network is trained to perform haplotype-agnostic and / or somatic variant calling. The method includes obtaining a reference sequence 310.

[0058] The reference sequence 310 may be obtained, for example, from one or more private or public repositories such as the Reference Sequence (RefSeq), an open-access, annotated and curated collection of nucleotide base sequences built by the National Center for Biotechnology Information (NCBI) or the NCBI Genomes FTP site storing a set of complete genomes of different organisms. It should be understood that in some embodiments, a specific copy of the reference sequence may be stored locally, while in other embodiments the reference sequence may be obtained from a remote server, e.g., through a computer network. Furthermore, in some embodiments, the entire reference sequence may be obtained while, in other embodiments, one or more sections of the reference sequence may be obtained - e.g., only the section(s) that is / are associated with a particular assay. Thus, a “reference sequence” as used herein refers generally to one or more sections of the reference sequence, which may or may not include the entire reference sequence.

[0059] In some embodiments, reference sequence data may include germline-associated data (e.g., multiple alternative reference sequences including germline variant data), particularly where a haplotype-aware approach is contemplated. In some embodiments, the reference sequence(s) omits some or substantially all germline variant information, particularly where a haplotype-agnostic approach is contemplated (e.g., where a candidate variant is identified as a likely somatic variant).

[0060] The haplotype-agnostic method includes obtaining a plurality of tumor sequence reads 311 and a plurality of corresponding normal (non-tumor) sequence reads 311-B. For example, in some embodiments, tumor sequence reads 311 can include sequencing results for a sample of tumor (cancerous) tissue of a patient, and normal sequence reads can include sequencing results for a sample of normal (non-cancerous) tissue taken (e.g., biopsied) from the same patient. In other embodiments, the normal sequence reads 311-B can include sequencing results for a sample of normal tissue taken from a different patient. Both types of sequence reads can be obtained, for example, by a sequencer instrument 90 in a manner similar to those described further herein.

[0061] A plurality of candidate somatic (and / or germline) variant positions can be obtained - either by obtaining pre-determined variant positions from a memory or a public / private database, a set of whitelist candidate positions, or by performing a comparison between the tumor sequence reads 311 and the reference sequence 310, for example, in the manner described above.

[0062] A segment of the normal sequence reads 311-B around a candidate variant position is converted into a normal sample matrix. Similarly, a segment of the tumor sequence reads 311 around the candidate variant position (substantially corresponding in length and location to the segment of normal sequence reads 311-B) is converted into a tumor and / or germline sample matrix 331-A. A segment of the reference sequence 310 around the candidate variant position (substantially corresponding in length and location to the segment of normal sequence reads) is converted into a reference matrix 330.

[0063] In some embodiments, before the above segments are converted into their respective reference matrices, each of them can be augmented using an augmentation technique. In some embodiments, this includes scanning the plurality of obtained sequence reads, comparing them to the obtained reference sequence 310, and determining a set of oneor more positions within the sequence reads that appear to include variants of some type (e.g., insertion or deletion variants (INDELs), single-nucleotide variants (SNVs), or structural variants (SVs)).

[0064] The sequence reads 311 & 311-B and the reference sequence 310 around the current candidate variant position may be augmented to achieve a precise multiple sequence alignment (MSA). Fig. 3 shows a plurality of exemplary sequence reads 311 & 311-B, an exemplary reference sequence 310, and an exemplary candidate variant position 350. As illustrated in FIG. 3, augmenting reference sequence 310 and sequence reads 311 & 311-B may include inserting one or more gaps or spaces (e.g., within reference sequence 310 and / or within one or more sequence reads 311 & 311-B, thereby producing augmented reference, tumor, and normal matrices 330, 331 -A and 331-B aligned with each other. In yet other embodiments, the augmentation step can be omitted altogether, and the original, unprocessed sequence reads 311 & 311-B and reference sequence 310 can be converted into their respective matrices directly.

[0065] The three matrices 300, 331 -A, & 331-B are fed (provided as inputs) into a trained neural network, and an output of the trained neural network is obtained, where the output includes at least a prediction (a probabilistic estimate) of the somatic or germline variant type contained in the tumor sequence read 311 (e.g., at the candidate somatic / germline variant position 350 or its vicinity). In some embodiments, the reference matrix 330 or normal sample matrices 331-B are omitted, such as in order to speed processing where a candidate variant is identified by the classifier model to not likely to require use of the reference / normal sequence information in order to accurately identify the candidate variant.

[0066] Further features pertaining to variant calling (haplotype-aware or haplotypeagnostic) known to those of ordinary skill in the art may be applied, such as further described in U.S. Patent Application Publication No. US-2021-0257050-A1, entitled “Systems and Methods for using Neural Networks for Germline and Somatic Variant Calling,” the entire contents of which is herein incorporated by reference.

[0067] FIG. 4 illustrates a process flow (i.e., a method) for a haplotype- aware variant calling process, in accordance with some embodiments. The method includes training a machine learning model 400 (e.g., one or more deep learning neural network model(s)) with haplotype-related sequence data. The data may include, for example, long-range haplotypedata 420 (e.g., based on longer than typical linkage disequilibrium to identify regions with historic recombination), multi-pseudogene region data 430 (e.g., including highly homologous and repetitive areas that may include high sequence similarity with known genes), distant allelles 440 (e.g., on different chromosomes or far apart on same chromosome), homogeneous regions 410 (e.g., within the human genome that may typically be difficult to distinguish or segment), and / or repetitive regions 450 (e.g., within the human genome that may also be difficult to distinguish from known variants).

[0068] This training data is used to train the model(s) to distinguish between germline variants of different types, between variants and non-variants or likely errors in sequencing data, for example. In some embodiments the model(s) perform better using multiple layers of nodes in order to process the complexity in the reference and sequencing data.

[0069] At block 460, a set of input sequencing data (e.g., sequenced from a patient sample) is fed into the model(s) 400 and processed by the model(s) (trained as further described above) to determine a likely variant and output a report at block 470. The output variant identification may include information about the likely variant such as further described herein, including structural properties, prediction quality, or an indication that no variant was detected and / or a likely sequencing error occurred. The report may include structural information and / or other properties (e.g., germline, somatic, etc.) about the predicted variant.

[0070] FIG. 5 illustrates a neural network 500, in accordance with at least some embodiments. The neural network 500 includes a plurality of serially connected convolutional layers 510 (510A-510I) eventually feeding into a first fully connected layer 520A, the output of which feeds into four separate fully connected layers 520A-520C that output various predictions. Different convolutional layers 510 may have different filter sizes (e.g., 1x3, 3x3, or 5x5) and may be interconnected through additional processing layers (not explicitly shown), such as Rectifying Linear Units (ReLUs), pooling layers, batch normalization (BN) layers, etc. Furthermore, as illustrated in FIG. 5, in some embodiments, an input to a certain layer can sometimes be connected via an identity shortcut connection 515 to an output of a subsequent layer in order to help maintain the signal in deeper layers of the network.

[0071] In some embodiments, the neural network 500 can output one or more predictions associated with a candidate variant within a segment of augmented sequence reads (e.g., matrix 331-A as illustrated in FIG. 3). A prediction can be obtained at the output of a fully connected layer 520B and can indicate whether the candidate variant position corresponds to a variant and if so the type of that variant. More specifically, in some embodiments, the prediction can include at least four probability values associated with four possible types of variants: NONE (no variant, i.e., a false-positive call); SNP / SNV (single nucleotide polymorphism / variant); INS (insertion variant); and DEL (deletion variant). Based on these probability values it can be determined which type of variant is the most likely, as well as the confidence level of the particular prediction.

[0072] In some embodiments, the neural network 500 can also output (e.g., at the output of fully connected layer 520C) the predicted position of the variant within the segment of (augmented) sequence reads represented by sample matrix 331-A. In other embodiments, this output can be omitted and it can be assumed that the position of the variant is known (e.g., at the center of the segment) based on the manner in which the segment of sequence reads is selected for a given candidate variant, as described above.

[0073] In some embodiments, the neural network 500 can also output (e.g., at the output of fully connected layer 520D) a predicted length of the variant. For example, a length of “1” can be outputted if the predicted variant is one base long (e.g., a SNP variant or a one-base DEL or INS variant); a length of “2” can be outputted if the predicted variant is a two-base long DEL or INS variant; and so forth. In some embodiments, if the output indicates that a variant is 2 bases long or longer, a post processing step can resolve the exact sequence being inserted or deleted.

[0074] In some embodiments, the neural network 500 can also output (e.g., at the output of fully connected layer 520E) a predicted genotype associated with the variant, representing, for example, the probabilities with the variant being 1) a homozygous reference (a nonvariant); 2) a heterozygous variant (where only one of maternal or paternal copies has a variant); 3) a homozygous variant (where both copies have the same variant); or 4) other (where each copy has a different variant). It is appreciated that in other embodiments, some of these outputs can be omitted and / or additional outputs can be added.

[0075] Referring now to the network’s inputs, in addition to reference matrix 330 and sample matrix 331 -A discussed above, in some embodiments other inputs can be provided to further improve the accuracy of the predictions and / or to expand the amount of information that can be obtained at the network’s output. Additional inputs may include, for example, a variant position matrix, a coverage matrix, and one or more alignment feature matrices.

[0076] In some embodiments, all input matrices can be provided into the neural network 500 as one large three-dimensional matrix. The dimensions of the matrix can be, for example, 5 x s x k, where 5 corresponds to the number of row / columns (-, A, C, T, G); s corresponds to the length of the segment (e.g., 16); and k corresponds to the number of different two- dimensional (e.g., 5 x s) matrices described above. In some embodiments, k can be as high as 30, or even higher.

[0077] A variant position matrix may comprise a two-dimensional representation of the candidate variant position within a reference and sample matrices. For examples, the position matrix can have five rows (-, A, C, T, G) and each column can represent a position within the segment. For example, a variant position matrix may include one value (e.g., 1) in all rows in the column corresponding to the candidate variant position, and another value (e.g., 0) in all rows of all other columns.

[0078] A coverage matrix may represent the coverage or depth of the particular segment of the sequence reads. For example, in some embodiments, a coverage matrix may include the same value in all of its elements, the value representing the coverage / depth (e.g., the average coverage / depth) of the different reads within the segment. In other embodiments, the coverage matrix may include different values at different columns, each value representing the respective coverage / depth at each column. Alignment feature matrices can represent various metrics associated with the quality of the sequence reads and their alignment. Such metrics can include, for example, base qualities, mapping qualities, strand bias, clipping information, and so forth. These alignment features can help improve the prediction accuracy by providing more context to sequence reads, for instance, by informing the network about the quality of the sequence bases and / or of their mapping. Additional inputs to the neural network 500 (not illustrated for brevity) may include various data related to known variants. Such data can be obtained, for example, from public and / or private knowledge bases such as dbSNP, COSMIC, ExAC, etc.

[0079] It is appreciated that FIG. 5 shows only one exemplary configuration of the neural network 500 and that neural networks having other suitable configurations / architectures can be used to analyze sequence reads (represented by a sample matrix) and to predict the type of variant and its other characteristics, without departing from the scope and spirit of the present disclosure.

[0080] Such alternative configurations may be selected based on a candidate variant’s properties and / or target type (e.g., structure, position, somatic, germline) and can include fewer layers, additional layers, layers having different parameters, additional inputs or outputs, fewer inputs or outputs, and so forth adapted to the candidate variant / target prediction properties / quality). Moreover, in some embodiments, the neural network 500 may not be a convolutional neural network (CNN), but may instead be another type of a deep neural network (DNN), i.e., another type of artificial neural network (ANN) that has multiple layers between the input and output layers, without departing from the scope and spirit of the present disclosure.Variant Calling Network Training

[0081] It is appreciated by a person skilled in the art that before neural network 500 could start performing accurate germline or somatic variant calling, it first needs to be trained on germline or somatic training data (i.e., training sequences), respectively. The training can include, for example, performing analysis on a large number training sequences, but then also providing to the network “ground truth” data (e.g., actual known variant types and their positions, lengths, genotypes, etc.) to enable the network to gradually minimize its output errors by adjusting its trainable filters and other trainable parameters after every run, such as through the process known as “back propagation.”

[0082] In some embodiments, the neural network 500 can be trained on genomes having well characterized ground truth variants, such as the NA12878 genome. In some embodiments, instead or in addition to real genomes, various simulation based strategies can be used to train the network. For example, to train the network for germline variant calling, synthetic samples with a given set of variants can be simulated using the method described in Mu, J. C. et al., “VarSim: a high-fidelity simulation and validation framework for high- throughput genome sequencing with cancer applications,” Bioinformatics 31, 1469-1471 (2015), which is herein incorporated by reference in its entirety. As another example, to trainthe network for somatic variant calling, it can be fed normal samples in which random variants have been spiked, e.g., using the method described in Eving, A. et al., “Combining tumor genome simulation with crowdsourcing to benchmark somatic single- nucleotide- variant detection,” Nature methods (2015), which is herein incorporated by reference in its entirety. Alternatively or in addition, two different normal germline samples with different proportions can be mixed to generate synthetic tumor / normal training samples. As another example, virtual tumor / normal training samples with desired allele frequency distributions can be generated by switching reads between two germline samples at variant locations.

[0083] In some embodiments, to further improve the accuracy of the variant calling, the neural network 500 can also be provided one or more other callers’ features matrices at its input. These matrices can represent (e.g., summarize) one or more features obtained from one or more other (e.g., third-party) variant calling applications or algorithms that have already processed the same segment of sequence reads. Such features can include variant types, lengths, and positions predicted by other applications / algorithms, as well as quality scores, mapping scores, variant significance scores, and other alignment features. In some embodiments, all input matrices being fed into the neural network 500 can be “combined” and provided as one large three-dimensional matrix.Exemplary Computer System

[0084] FIG. 6 illustrates an example computer system that may be utilized to implement techniques disclosed herein. Any of the computer systems mentioned herein, such as for hosting the systems and implementing the processes described for calculating variants, may utilize any suitable number of subsystems. Examples of such subsystems are shown in FIG. 6 as included in computer system 10. In some embodiments, a computer system includes a single computer apparatus, where the subsystems can be the components of the computer apparatus. In other embodiments, a computer system can include multiple computer apparatuses, each being a subsystem, with internal components. A computer system can include desktop and laptop computers, tablets, mobile phones, telecommunication devices or other mobile devices. In some embodiments, a cloud infrastructure (e.g., Amazon Web Services), a graphical processing unit (GPU), etc., can be used to implement the disclosed techniques.

[0085] The subsystems shown in FIG. 6 are interconnected via a system bus 75. Additional subsystems such as a printer 74, keyboard 78, storage device(s) 79, monitor 76, which is coupled to display adapter 82, and others are shown. Peripherals and input / output (I / O) devices, which couple to I / O controller 71 , can be connected to the computer system by any number of means known in the art such as input / output (I / O) port 77 (e.g., USB, FireWire®). For example, I / O port 77 or external interface 81 (e.g. Ethernet, Wi-Fi, etc.) can be used to connect computer system 10 to a wide area network such as the Internet, a mouse input device, or a scanner.

[0086] The interconnection via system bus 75 allows the central processor 73 to communicate with each subsystem and to control the execution of a plurality of instructions from system memory 72 or the storage device(s) 79 (e.g., a fixed disk, such as a hard drive, or optical disk), as well as the exchange of information between subsystems. The system memory 72 and / or the storage device(s) 79 may embody a computer readable medium. Another subsystem is a data collection device 85, such as a camera, microphone, accelerometer, and the like. Any of the data mentioned herein can be output from one component to another component and can be output to the user.

[0087] A sequencer instrument 90 (e.g., a nanopore sequencer), which may sometimes be referred to as a sequencing device, is connected through external interface 81 for providing sequencing data to a data collection device 85 and / or storage devices 79.

[0088] The sequencer instrument 90 may be any sequencer instrument including, but not limited to, the following: sequencing-by-hybridization platforms from Affymetrix, Inc., now ThermoFisher Scientific, (Sunnyvale, Calif.), sequencing-by-synthesis platforms from Illumina, Inc. (San Diego, Calif.) and Helicos Biosciences (Cambridge, Mass.), and sequencing-by-ligation platform from Applied Biosystems (Foster City, Calif.). Other sequencing technologies include, but are not limited to, the Ion Torrent technology from ThermoFisher Scientific (Waltham, MA); as well as nanopore sequencing from Roche Sequencing Solutions, Inc. (Santa Clara, Cal.) and Oxford Nanopore Technologies (Oxford, United Kingdom).

[0089] A computer system 10 can include a plurality of the same components or subsystems, e.g., connected together by external interface 81 or by an internal interface such as the bus 75. In some embodiments, computer systems, subsystems, or apparatuses cancommunicate over a network. In such instances, one computer can be considered a client and another computer a server, where each can be part of the same computer system. A client and a server can each include multiple systems, subsystems, or components.

[0090] Aspects of embodiments can be implemented in the form of control logic using hardware (e.g. an application specific integrated circuit or field programmable gate array) and / or using computer software with a generally programmable processor in a modular or integrated manner. As used herein, a processor includes a single-core processor, multi-core processor on a same integrated chip, or multiple processing units on a single circuit board or networked. Based on the disclosure and teachings provided herein, a person of ordinary skill in the art will know and appreciate other ways and / or methods to implement embodiments of the present invention using hardware and a combination of hardware and software.

[0091] Machine learning models utilized herein may include one or more of a Naive Bayes (NB) model, a logistic regression (LR) model, a random forest (RF) model, a support vector machine (SVM) model, an artificial neural network model, a multilayer perceptron (MLP) model, a convolutional neural network (CNN), a Large Language model (LLM), and / or other machine learning or deep learning models, etc. The machine learning models can be updated / trained using a supervised learning technique, an unsupervised learning technique, etc.

[0092] Any of the software components or functions described in this application may be implemented as software code to be executed by a processor using any suitable computer language such as, for example, Java, C, C++, C#, Objective-C, Swift, or scripting language such as Perl or Python using, for example, conventional or object-oriented techniques. The software code may be stored as a series of instructions or commands on a computer readable medium for storage and / or transmission. A suitable non-transitory computer readable medium can include random access memory (RAM), a read only memory (ROM), a magnetic medium such as a hard-drive or a floppy disk, or an optical medium such as a compact disk (CD) or DVD (digital versatile disk), flash memory, and the like. The computer readable medium may be any combination of such storage or transmission devices.

[0093] Such programs may also be encoded and transmitted using carrier signals adapted for transmission via wired, optical, and / or wireless networks conforming to a variety of protocols, including the Internet. As such, a computer readable medium may be created usinga data signal encoded with such programs. Computer readable media encoded with the program code may be packaged with a compatible device or provided separately from other devices (e.g., via Internet download). Any such computer readable medium may reside on or within a single computer product (e.g. a hard drive, a CD, or an entire computer system), and may be present on or within different computer products within a system or network. A computer system may include a monitor, printer, or other suitable display for providing any of the results mentioned herein to a user.

[0094] Any of the methods described herein may be totally or partially performed with a computer system including one or more processors, which can be configured to perform the steps. Thus, embodiments can be directed to computer systems configured to perform the steps of any of the methods described herein, potentially with different components performing a respective step or a respective group of steps. Although presented as numbered steps, steps of methods herein can be performed at the same time or in a different order. Additionally, portions of these steps may be used with portions of other steps from other methods. Also, all or portions of a step may be optional. Additionally, any of the steps of any of the methods can be performed with modules, units, circuits, or other means for performing these steps.

[0095] The specific details of particular embodiments may be combined in any suitable manner without departing from the spirit and scope of embodiments of the invention. However, other embodiments of the invention may be directed to specific embodiments relating to each individual aspect, or specific combinations of these individual aspects.

[0096] The above description of example embodiments of the invention has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form described, and many modifications and variations are possible in light of the teaching above.

[0097] A recitation of "a", "an" or "the" is intended to mean "one or more" unless specifically indicated to the contrary. The use of “or” is intended to mean an “inclusive or,” and not an “exclusive or” unless specifically indicated to the contrary. Reference to a “first” component does not necessarily require that a second component be provided. Moreover, reference to a “first” or a “second” component does not limit the referenced component to a particular location unless expressly stated.

[0098] All patents, patent applications, publications, and descriptions mentioned herein are incorporated by reference in their entirety for all purposes. None is admitted to be prior art.

Claims

CLAIMS1. A computer-implemented method for identifying a genomic variant, the method comprising: obtaining one or more reference sequences; obtaining sequencing data comprising a plurality of sequence reads pertaining to a biological sample; determining one or more candidate variant positions from the plurality of sequence reads; for each candidate variant position, applying a classifier model for selecting use of a haplotype-aware or haplotype- agnostic variant analysis with respect to the candidate variant position; based on the output of the classifier model for each candidate variant position, respectively applying a haplotype- aware or haplotype-agnostic variant analysis for the corresponding candidate variant position; and transmitting an output variant identification for each candidate variant position based on the respectively applied haplotype-aware or haplotype-agnostic variant analysis.

2. The method of claim 1, wherein the classifier model is trained from sequencing data identified with somatic variants in a plurality of regions.

3. The method of claim 1, wherein the classifier model is trained using performance measures for identifying variants based on one or more candidate regions of interest.

4. The method of claim 1 , wherein at least one of the haplotype-aware or haplotypeagnostic variant analysis comprises applying a neural network trained with sequencing data identified with variants in a plurality of regions.

5. The method of claim 1, wherein the haplotype-aware variant analysis comprises: obtaining augmented sequence reads by inserting one or more spaces in the sequencing data; obtaining an augmented reference sequence by inserting one or more spaces in the one or more reference sequences;converting a segment of the augmented sequence reads around the respective candidate variant into a sample matrix; converting a segment of the augmented reference sequence around the respective candidate variant into a reference matrix; providing the sample matrix and the reference matrix to a trained neural network; and receiving, at the output of the trained neural network, prediction data relating to one or more variants identified within the sequencing data.

6. The method of claim 5, wherein augmenting the sequence reads and the reference sequence comprises: detecting one or more inserted bases in the sequencing data; and for each inserted base detected, inserting a space in the reference sequence at a position in the reference sequence corresponding to a position of the inserted base, and inserting a space at the position of the inserted base in every sequence read of the sequencing data for which no insertions were detected at the position of the inserted base.

7. The method of claim 5, wherein the trained neural network is provided with at least one of: a variant position matrix representing a respective candidate variant’s position within the segment of the augmented sequence reads; a coverage matrix representing coverage or depth of the segment of the augmented sequence reads; an alignment feature matrix representing an alignment feature of the augmented sequence reads; or a knowledgeable base matrix representing predetermined information about one or more variants.

8. The method of claim 1, wherein the sequencing data comprises: sequence reads from a normal sample of an individual, and sequence reads from a tumor sample of the individual, and wherein the haplotype-agnostic variant analysis comprises: converting a segment of the sequence reads for the normal sample and a segment of the sequence reads for the tumor sample into a normal sample matrix and a tumor sample matrix, respectively;feeding the normal sample matrix and the tumor sample matrix into a trained convolutional neural network.

9. The method of claim 8, wherein the sequence reads for the tumor sample represent genetic information of tissue from a tumor biopsy.

10. The method of claim 8, wherein: converting the segment of the sequence reads for the normal sample into the normal sample matrix comprises augmenting the segment of the sequence reads for the normal sample by inserting one or more spaces in one or more sequence reads for the normal sample; and converting the segment of the sequence reads for the tumor sample into the tumor sample matrix comprises augmenting the segment of the sequence reads for the tumor sample by inserting one or more spaces in one or more sequence reads for the tumor sample.

11. A system for identifying a genomic variant in a biological sample, the system comprising: one or more processors configured to: obtain one or more reference molecular sequences; obtain sequence data comprising a plurality of sequence reads pertaining to the biological sample; determine one or more candidate variant positions from the plurality of sequence reads; for each candidate variant position, apply a classifier model for selecting use of a haplotype- aware or haplotype-agnostic variant analysis with respect to the candidate variant position; based on the output of the classifier model, respectively apply a haplotype- aware or haplotype-agnostic variant analysis for each candidate variant position; and transmit an output variant identification for each candidate variant position based on the respectively applied haplotype-aware or haplotype-agnostic variant analysis.

12. The system of claim 11, wherein the classifier model is trained from sequencing data identified with somatic variants in a plurality of regions.

13. The system of claim 11, wherein the classifier model is trained using performance measures for identifying variants based on one or more candidate regions of interest.

14. The system of claim 11, wherein at least one of the haplotype-aware or haplotypeagnostic variant analysis comprises applying a neural network trained with sequencing data identified with variants in a plurality of regions.

15. The system of claim 11, wherein the haplotype-aware variant analysis comprises: obtaining augmented sequence reads by inserting one or more spaces in the sequencing data; obtaining an augmented reference sequence by inserting one or more spaces in the one or more reference sequences; converting a segment of the augmented sequence reads around the respective candidate variant into a sample matrix; converting a segment of the augmented reference sequence around the respective candidate variant into a reference matrix; providing the sample matrix and the reference matrix to a trained neural network; and receiving, at the output of the trained neural network, prediction data relating to one or more variants identified within the sequencing data.

16. The system of claim 15, wherein augmenting the sequence reads and the reference sequence comprises: detecting one or more inserted bases in the sequencing data; and for each inserted base detected, inserting a space in the reference sequence at a position in the reference sequence corresponding to a position of the inserted base, and inserting a space at the position of the inserted base in every sequence read of the sequencing data for which no insertions were detected at the position of the inserted base.

17. The system of claim 15, wherein the trained neural network is provided with at least one of:a variant position matrix representing a respective candidate variant’s position within the segment of the augmented sequence reads; a coverage matrix representing coverage or depth of the segment of the augmented sequence reads; an alignment feature matrix representing an alignment feature of the augmented sequence reads; or a knowledgeable base matrix representing predetermined information about one or more variants.

18. The system of claim 11, wherein the sequencing data comprises: sequence reads from a normal sample of an individual, and sequence reads from a tumor sample of the individual, and wherein the haplotype-agnostic variant analysis comprises: converting a segment of the sequence reads for the normal sample and a segment of the sequence reads for the tumor sample into a normal sample matrix and a tumor sample matrix, respectively; and feeding the normal sample matrix and the tumor sample matrix into a trained convolutional neural network.

19. The system of claim 18, wherein the sequence reads for the tumor sample represent genetic information of tissue from a tumor biopsy.

20. The system of claim 18, wherein: converting the segment of the sequence reads for the normal sample into the normal sample matrix comprises augmenting the segment of the sequence reads for the normal sample by inserting one or more spaces in one or more sequence reads for the normal sample; and converting the segment of the sequence reads for the tumor sample into the tumor sample matrix comprises augmenting the segment of the sequence reads for the tumor sample by inserting one or more spaces in one or more sequence reads for the tumor sample.

Images