# US 12,763,163 B2 · Cancer classification with genomic region modeling > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,763,163 B2 - Document ID: US12763163B2 - Status: Granted - Kind: B2 - Application number: 17/216,551 - Filed: 2021-03-29 - Published: 2021-10-07 - Granted: 2026-10-13 - Earliest claimed: 2020-03-31 - Assignees: Grail, LLC - Applicants: Grail, Inc. - Inventors: Samuel S. Gross; Joshua Newman; Virgil Nicula - Web page: https://patentgrep.com/patent/US12763163B2 ## CPC - G16B 5/20: Probabilistic models - G16B 40/00: ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding - G16H 70/60: relating to pathologies - G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Methods and systems for detecting cancer and/or determining a cancer tissue of origin are disclosed. Fragments are grouped into genomic regions, wherein a region model is trained for each genomic region. Fragments are input into the region models, and the outputs are used to generate a feature vector for cancer classification. In one embodiment, the region models are shallow neural networks configured to generate a score indicating a likelihood that a fragment is derived from a cancer biological sample. The feature vector is determined based on counts of fragments having scores above threshold scores for the various genomic regions. In another embodiment, the regions models are configured to generate a region embedding for an input methylation embedding of a fragment. The region embeddings are pooled by region and then pooled again to generate the feature vector. ## Claims ### Claim 1 A method comprising: performing, by a sequencing device, a sequencing assay with a targeted panel targeting a plurality of genomic regions, wherein the targeted panel includes, for each genomic region, one or more hybridization probes for capturing cfDNA fragments originating from the genomic region, wherein the sequencing assays yields sequencing data for a biological sample comprising a plurality of sequence reads representing cfDNA fragments, each sequence read overlapping at least one genomic region of a plurality of genomic regions; for each cfDNA fragment of the biological sample, determining, by a computer processor, a first score for the genomic region that the cfDNA fragment overlaps, the first score for a genomic region determined by inputting the sequence read for the cfDNA fragment into a shallow neural network trained for the genomic region, the shallow neural network configured to generate the first score representative of a likelihood that the cfDNA fragment is derived from a cancer biological sample, each shallow neural network of a plurality of shallow neural networks spanning the plurality of genomic regions comprises no more than three hidden layers and is trained by: accessing a first plurality of sequence reads obtained from cancer samples and a second plurality of sequence reads obtained from non-cancer samples, wherein the first plurality of sequence reads and the second plurality of sequence reads overlap the genomic region, and training the shallow neural network with the first plurality of sequence reads and the second plurality of sequence reads; generating, by the computer processor, a feature vector for the biological sample, each feature of the feature vector corresponding to a genomic region of the plurality of genomic regions and generated according to a count of cfDNA fragments having a score for the genomic region above a threshold score; and inputting, by the computer processor, the feature vector into a trained machine-learning model to generate a cancer prediction for the biological sample, the trained model trained by: accessing sequencing data for a plurality of cancer samples and a plurality of non-cancer samples, generating a feature vector for each cancer sample and for each non-cancer sample by applying the shallow neural networks to the sequencing data of each sample, and training the trained model with the feature vectors for the cancer samples and the non-cancer samples. ### Claim 2 (depends on 1) The method of claim 1, wherein each shallow neural network consists of one hidden layer. ### Claim 3 (depends on 2) The method of claim 2, wherein the hidden layer in each shallow neural network comprises no more than one of: 8 nodes, 9 nodes, 10 nodes, 11 nodes, 12 nodes, 16 nodes, 20 nodes, 24 nodes, 28 nodes, and 32 nodes. ### Claim 4 (depends on 1) The method of claim 1, wherein each shallow neural network consists of two hidden layers. ### Claim 5 (depends on 1) The method of claim 1, wherein a first genomic region comprises a first number of CpG sites and a second genomic region in the plurality of genomic regions comprises a second number of CpG sites that is different than the first number of CpG sites. ### Claim 6 (depends on 1) The method of claim 1, wherein each shallow neural network outputs the first score that corresponds to a likelihood that a cfDNA fragment is derived from a biological sample of a first cancer type and a second score that corresponds to a likelihood that the cfDNA fragment is derived from a biological sample of a second cancer type different than the first cancer type. ### Claim 7 (depends on 1) The method of claim 1, wherein each feature of the feature vector is generated according to a normalization of the count of cfDNA fragments having a score for the genomic region above the threshold score. ### Claim 8 (depends on 1) The method of claim 1, wherein each cfDNA fragment is an anomalous fragment, the method further comprising: filtering an initial set of cfDNA fragments with p-value filtering to generate the set of anomalous fragments, the filtering comprising removing fragments from the initial set having below a threshold p-value with respect to other fragments to produce the set of anomalous fragments. ### Claim 9 (depends on 1) The method of claim 1, wherein the trained machine-learning model is a neural network algorithm, a support vector machine algorithm, a Naive Bayes algorithm, a nearest neighbor algorithm, a boosted trees algorithm, a random forest algorithm, a decision tree algorithm, a multinomial logistic regression algorithm, a linear model, or a linear regression algorithm. ### Claim 10 A method for training a machine-learning cancer prediction model, comprising: performing, by a sequencing device on a plurality of cancer samples and a plurality of non-cancer samples, a sequencing assay with a targeted panel targeting a plurality of genomic regions, wherein the targeted panel includes, for each genomic region, one or more hybridization probes for capturing cfDNA fragments originating from the genomic region, wherein the sequencing assays yields sequencing data for each sample comprising a plurality of sequence reads representing the cfDNA fragments, with each sequence read overlapping at least one genomic region of a plurality of genomic regions; for each sequence read in each sample, determining, by a computer processor, a first score for the genomic region that the sequence read overlaps, the first score for a genomic region determined by inputting the sequence read into a shallow neural network trained for the genomic region, the shallow neural network configured to generate the first score representative of a likelihood that the sequence read is derived from a cancer biological sample, each shallow neural network of a plurality of shallow neural networks spanning the plurality of genomic regions comprises no more than three hidden layers and is trained by: accessing a first plurality of sequence reads obtained from cancer samples and a second plurality of sequence reads obtained from non-cancer samples, wherein the first plurality of sequence reads and the second plurality of sequence reads overlap the genomic region, and training the shallow neural network with the first plurality of sequence reads and the second plurality of sequence reads; generating, by the computer processor, a feature vector for each sample, each feature of the feature vector corresponding to a genomic region of the plurality of genomic regions and generated according to a count of sequence reads having a score for the genomic region above a threshold score; and training, by the computer processor, the machine-learning cancer prediction model with the feature vectors for the cancer samples and the non-cancer samples, wherein the trained cancer prediction model is configured to generate a cancer prediction for a biological sample. ### Claim 11 (depends on 10) The method of claim 10, wherein each shallow neural network consists of one hidden layer. ### Claim 12 (depends on 11) The method of claim 11, wherein each hidden layer in each shallow neural network comprises no more than one of: 8 nodes, 9 nodes, 10 nodes, 11 nodes, 12 nodes, 16 nodes, 20 nodes, 24 nodes, 28 nodes, and 32 nodes. ### Claim 13 (depends on 10) The method of claim 10, wherein a first genomic region comprises a first number of CpG sites and a second genomic region in the plurality of genomic regions comprises a second number of CpG sites that is different than the first number of CpG sites. ### Claim 14 (depends on 10) The method of claim 10, wherein each shallow neural network outputs the first score that corresponds to a likelihood that the sequence read is derived from a biological sample of a first cancer type and a second score that corresponds to a likelihood that the sequence read is derived from a biological sample of a second cancer type different than the first cancer type. ### Claim 15 (depends on 10) The method of claim 10, wherein each feature of the feature vector is generated according to a normalization of the count of sequence reads having a score for the genomic region above the threshold score.