# US 11,443,832 B2 · Genetic mutation detection using deep learning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,832 B2 - Document ID: US11443832B2 - Status: Granted - Kind: B2 - Application number: 16/296,135 - Filed: 2019-03-07 - Published: 2020-09-10 - Granted: 2022-09-13 - Earliest claimed: 2019-03-07 - Assignees: NVIDIA Corporation - Applicants: NVIDIA Corporation - Inventors: Johnny Israeli; Avantika Lal; Michael Vella; Nikolai Yakovenko; Zhen Hu - Web page: https://patentgrep.com/patent/US11443832B2 ## CPC - G16B 40/30: Unsupervised data analysis - G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting - G06F 18/24: Classification techniques - G06N 3/0418: {using chaos or fractal principles} - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - G06N 3/096: Transfer learning - G16B 20/20: Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection - G16B 30/00: ICT specially adapted for sequence analysis involving nucleotides or amino acids - G16B 40/20: Supervised data analysis - G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract The present disclosure provides methods, systems, and computer program products that use deep learning models to classify candidate mutations detected in sequencing data, particularly suboptimal sequencing data. The methods, systems, and programs provide for increased efficiency, accuracy, and speed in identifying mutations from a wide range of sequencing data. ## Claims ### Claim 1 A method comprising: encoding information representing a candidate mutation, wherein the information comprises absolute location, reference sequence, number of reads (N), and read sequences, wherein the reference sequence and the read sequences are encoded as 2D Boolean matrices and the reference sequence is associated with each of the read sequences; and processing the encoded information with a deep learning model that jointly processes the reference sequence with each read sequence and classifies the candidate mutation. ### Claim 2 (depends on 1) The method of claim 1, wherein the encoding augments the reference sequence. ### Claim 3 (depends on 1) The method of claim 1, wherein the encoded information further comprises read mapping quality values and/or read base quality values. ### Claim 4 (depends on 1) The method of claim 1, wherein processing with a deep learning model comprises transforming the encoded information in parallel with a plurality of CNNs of different kernel sizes and combining the output into a tensor. ### Claim 5 (depends on 4) The method of claim 4, wherein combining the output comprises generating a ranking score and using it to generate TopK-selected tensor. ### Claim 6 (depends on 4) The method of claim 4, wherein the processing with a deep learning model further comprises transforming the tensor with a CNN comprising a 1D convolutional layer followed by a plurality of fully connected layers, wherein the final fully connected layer outputs the classification of the candidate mutation. ### Claim 7 (depends on 1) The method of claim 1, wherein the method further comprises training the deep learning model, wherein training comprises: encoding information representing a suboptimal candidate mutation, wherein the encoded information comprises absolute location, a model reference sequence, number of reads (N), and read sequences; processing the encoded information with a deep learning model that classifies the suboptimal candidate mutation; and minimizing error in the classification of the suboptimal candidate mutation relative to a ground truth candidate mutation of the model reference sequence by adjusting parameters of the deep learning model. ### Claim 8 (depends on 7) The method of claim 7, wherein the encoding associates the reference sequence with each read sequence, whereby the deep learning model jointly processes the reference sequence with each read sequence. ### Claim 9 (depends on 8) The method of claim 8, wherein processing with a deep learning model comprises transforming the encoded information in parallel with a plurality of CNNs of different kernel sizes and combining the output into a tensor. ### Claim 10 (depends on 9) The method of claim 9, wherein the processing with a deep learning model further comprises transforming the tensor with a CNN comprising a 1D convolutional layer followed by a plurality of fully connected layers, wherein the final fully connected layer outputs the classification of the suboptimal candidate mutation. ### Claim 11 A system comprising a processor, a memory device, and a classification engine executable on the processor according to software instructions stored in the memory device, wherein the classification engine is configured to: encode information representing a candidate mutation, wherein the encoded information comprises absolute location, reference sequence, number of reads (N), and read sequences, wherein the reference sequence and the read sequences are encoded as 2D Boolean matrices and the reference sequence is associated with each of the read sequences; and process the encoded information with a deep learning model that jointly processes the reference sequence with each read sequence and classifies the candidate mutation. ### Claim 12 (depends on 11) The system of claim 11, wherein the encoded information further comprises read mapping quality values and/or read base quality values. ### Claim 13 (depends on 11) The system of claim 11, wherein the system further comprises a training engine executable on the processor according to software instructions stored in the memory device, wherein the training engine is configured to: encode information representing a suboptimal candidate mutation, wherein the encoded information comprises absolute location, a model reference sequence, number of reads (N), and read sequences, wherein the reference sequence and the read sequences are encoded as 2D Boolean matrices and the reference sequence is associated with each of the read sequences; process the encoded information with a deep learning model that jointly processes the reference sequence with each read sequence and classifies the suboptimal candidate mutation; and minimize error in the classification of the suboptimal candidate mutation relative to a ground truth candidate mutation of the model reference sequence by adjusting parameters of the deep learning model. ### Claim 14 A non-transitory computer-readable medium comprising instructions for classifying candidate mutations that, when executed by a processor, cause the processor to perform one or more steps comprising: encoding information representing a candidate mutation, wherein the encoded information comprises absolute location, reference sequence, number of reads (N), and read sequences, wherein the reference sequence and the read sequences are encoded as 2D Boolean matrices and the reference sequence is associated with each of the read sequences; and processing the encoded information with a deep learning model that jointly processes the reference sequence with each read sequence and classifies the candidate mutation. ### Claim 15 (depends on 14) The non-transitory computer-readable medium of claim 14, wherein the medium further comprises instructions for training the deep learning model, wherein training comprises: encoding information representing a suboptimal candidate mutation, wherein the encoded information comprises absolute location, a model reference sequence, number of reads (N), and read sequences, wherein the reference sequence and the read sequences are encoded as 2D Boolean matrices and the reference sequence is associated with each of the read sequences; processing the encoded information with a deep learning model that jointly processes the reference sequence with each read sequence and classifies the suboptimal candidate mutation; and minimizing error in the classification of the suboptimal candidate mutation relative to a ground truth candidate mutation of the model reference sequence by adjusting parameters of the deep learning model.