# US 12,014,831 B2 · Approaches to reducing dimensionality of genetic information used for machine learning and systems for implementing the same > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,014,831 B2 - Document ID: US12014831B2 - Status: Granted - Kind: B2 - Application number: 18/073,471 - Filed: 2022-12-01 - Published: 2023-10-19 - Granted: 2024-06-18 - Earliest claimed: 2021-12-02 - Assignees: AIONCO, Inc. - Applicants: AIONCO, Inc. - Inventors: Cheuk Ying Tang; Victor Solovyev; Gene Lee - Web page: https://patentgrep.com/patent/US12014831B2 ## CPC - G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems - G06N 3/08: Learning methods - G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound - G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks - G06N 20/10: using kernel methods, e.g. support vector machines [SVM] - G16B 20/20: Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection - G16B 30/10: Sequence alignment; Homology search - G16B 40/20: Supervised data analysis - G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY ## Abstract Introduced here is an approach to further refining a set of locations that can serve as inputs to a machine learning algorithm. These locations may refer to unique molecular positions in a reference human genome and/or unique mutations thereof that are relevant in diagnosing cancer. A computing system can determine the diagnostic relevance of each location and then discard some of the less diagnostically relevant locations. ## Claims ### Claim 1 A method comprising: receiving an input indicative of an instruction to train a neural network to identify text phrases that are representative of mutations that are diagnostically relevant for a given type of cancer, wherein each of the text phrases is representative of a different set of characters, each of which is representative of a nucleotide; accessing a dataset that includes genetic information of multiple individuals that are known to have the given type of cancer; generating a first set of locations by examining the dataset, wherein each of the locations included in the first set is representative of a different molecular position at which a mutation is discovered through analysis of the genetic information; producing a set of metrics by computing, for each of the locations included in the first set, a metric that is indicative of correlation with the given type of cancer, as determined based on an analysis of the genetic information of the multiple individuals; generating a second set that includes fewer locations than the first set by identifying the locations included in the first set in order based on the set of metrics, ordered from most correlated with the given type of cancer to least correlated with the given type of cancer, and filtering at least some of the locations included in the first set, so as to produce the second set; and training the neural network using the second set. ### Claim 2 (depends on 1) The method of claim 1, wherein generating of the second set further comprises: implementing a clustering algorithm to discover groupings of the locations, identifying a first grouping and a second grouping that are separated by at least a predetermined amount, wherein the first grouping includes locations that are higher in the order than locations in the second grouping, and discarding (i) the locations included in the second grouping and (ii) locations included in groupings, if any, that correlate with the given type of cancer less than the second grouping. ### Claim 3 (depends on 1) The method of claim 1, wherein said generating of the second set comprises: discarding the locations whose relevance metrics do not exceed a threshold. ### Claim 4 (depends on 1) The method of claim 1, wherein said generating of the second set comprises: identifying a highest N number of the locations based on an analysis of the order, and discarding the locations that are not part of the highest N number. ### Claim 5 (depends on 1) The method of claim 1, wherein said producing comprises: for each of the locations, outputting a score indicative of importance in producing a diagnostically accurate output. ### Claim 6 (depends on 5) The method of claim 5, wherein each score is representative of the metric produced for the corresponding location. ### Claim 7 (depends on 5) The method of claim 5, wherein said producing further comprises: for each of the locations, determining the corresponding metric based on the corresponding score output by the neural network. ### Claim 8 (depends on 5) The method of claim 5, wherein the score is indicative of accuracy, precision, recall, sensitivity, specificity, or localization. ### Claim 9 (depends on 1) The method of claim 1, wherein said producing comprises: for each of the locations, outputting multiple scores that are indicative of different performance metrics, and determining the corresponding metric based on the multiple scores. ### Claim 10 (depends on 1) The method of claim 1, further comprising: initiating a connection with a network-accessible source of genomic data; and causing the dataset to be downloaded from the network-accessible source. ### Claim 11 (depends on 10) The method of claim 10, wherein said initiating and said causing are performed in response to said receiving. ### Claim 12 A non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, cause the computing device to perform operations comprising: receiving input indicative of an instruction to train a machine learning (ML) algorithm to produce a model that is able to identify mutations that are indicative of a given type of cancer; accessing genomic data associated with one or more individuals that are known to have the given type of cancer; generating a first list of locations by examining the genomic data, wherein each location included in the first list is representative of a different molecular position at which a mutation is discovered through analysis of the genomic data; reducing the first list through systematic analysis of correlation of each location with the given type of cancer, as determined based on an analysis of the genomic data, so as to produce a second list that has fewer locations than the first list; and storing the second list in a data structure that is associated with the given type of cancer. ### Claim 13 (depends on 12) The non-transitory medium of claim 12, wherein the genomic data is associated with a single patient that is known to have the given type of cancer. ### Claim 14 (depends on 12) The non-transitory medium of claim 12, wherein the genomic data is associated with multiple patients that are known to have the given type of cancer. ### Claim 15 (depends on 12) The non-transitory medium of claim 12, wherein the operations further comprise: downloading the genomic data from a repository that is manipulable via a data portal accessible via the Internet. ### Claim 16 (depends on 15) The non-transitory medium of claim 15, wherein the repository is associated with the National Cancer Institute, and wherein the genomic data is representative of a The Cancer Genome Atlas (TCGA) dataset. ### Claim 17 (depends on 12) The non-transitory medium of claim 12, wherein the operations further comprise: training the ML algorithm using the second set, so as to produce the model. ### Claim 18 (depends on 17) The non-transitory medium of claim 17, wherein upon being applied to new genomic data associated with a patient whose health state is unknown, the model utilizes the locations included in the second list to determine whether any mutations are present that are indicative of the given type of cancer. ### Claim 19 (depends on 17) The non-transitory medium of claim 17, wherein the operations further comprise: receiving second input indicative of request to analyze new genomic data associated with a patient whose health state is unknown; applying the model to the new genomic data, so as to produce an output that indicates whether the patient is determined to have the given type of cancer; and causing display of a visual indicium of the output.