# US 12,191,000 B2 · Systems and methods for classifying patients with respect to multiple cancer classes > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,191,000 B2 - Document ID: US12191000B2 - Status: Granted - Kind: B2 - Application number: 18/151,197 - Filed: 2023-01-06 - Published: 2023-06-01 - Granted: 2025-01-07 - Earliest claimed: 2018-12-10 - Assignees: Grail, LLC - Applicants: Grail, LLC - Inventors: M. Cyrus Maher; Anton Valouev; Darya Filippova; Virgil Nicula; Karthik Jagadeesh; Oliver Claude Venn; Samuel S. Gross; John F. Beausang; Robert Abe Paine Calef - Web page: https://patentgrep.com/patent/US12191000B2 ## CPC - G16B 30/00: ICT specially adapted for sequence analysis involving nucleotides or amino acids - G16B 40/20: Supervised data analysis - G06N 5/04: Inference or reasoning models - G06N 20/00: Machine learning - G16B 20/00: ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations - G16B 20/20: Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection - 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 10/40: for data related to laboratory analysis, e.g. patient specimen analysis - G16H 10/60: for patient-specific data, e.g. for electronic patient records - G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems - G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients - 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 - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Technical solutions for classifying patients with respect to multiple cancer classes are provided. The classification can be done using cell-free whole genome sequencing information from subjects. A reference set of subjects is used to train classifiers to recognize genomic markers that distinguish such cancer classes. The classifier training includes dividing the reference genome into a set of non-overlapping bins, applying a dimensionality reduction method to obtain a feature set, and using the feature set to train classifiers. For subjects with unknown cancer class, the trained classifiers provide probabilities or likelihoods that the subject has a respective cancer class for each cancer in a set of cancer classes. The present disclosure thus describes methods to improve the screening and detection of cancer class from among several cancer classes. This serves to facilitate early and appropriate treatment for subjects afflicted with cancer. ## Claims ### Claim 1 A method of training an untrained first classifier to classify a test subject of a given species to a cancer class in a plurality of cancer classes using a computer system comprising one or more processors, the method comprising: obtaining, by the computer system and for each respective reference subject in a first plurality of reference subjects, (i) a cancer class of the respective reference subject and (ii) a sequencing construct for the respective reference subject that includes a first bin count for each respective bin in a plurality of bins that collectively represent all or a portion of a reference genome of the species, wherein each respective first bin count representative of a number of nucleic acid fragments measured from nucleic acids in a biological sample obtained from the respective reference subject that maps onto a different and non-overlapping portion of the reference genome of the species, wherein, for each respective cancer class in the plurality of cancer classes, the first plurality of reference subjects includes at least one reference subject that has the respective cancer class; improving a computational efficiency of the computer system by collectively subjecting, by the computer system, the first bin count of each bin in the plurality of bins for each reference subject in the first plurality of reference subjects to a dimensionality reduction method thereby obtaining a feature set, wherein the feature set consists of a number of features that is fewer than the number of bins in the plurality of bins; resampling, using the computer system, the feature set a plurality of times, wherein the resampling comprises forming, for each respective training iteration in a plurality of training iterations, a trained component classifier via: omitting, from the feature set, a subset of values for features in the feature set for the first plurality of reference subjects; and forming the trained component classifier by inputting, in conjunction with the cancer class of respective reference subjects in the first plurality of reference subjects as ground truth, remaining values for the features in the feature set as collective input to a respective untrained component classifier; and constructing, as a result of the resampling and by collectively leveraging output generated by the trained component classifier formed in each of the plurality of training iterations, a trained first classifier having an improved cancer class recognition ability over the untrained first classifier. ### Claim 2 (depends on 1) The method of claim 1, wherein the sequencing construct for each respective reference subject in the first plurality of reference subjects is obtained by targeted panel or whole genome sequencing. ### Claim 3 (depends on 1) The method of claim 1, wherein each respective reference subject in the first plurality of reference subjects is human. ### Claim 4 (depends on 1) The method of claim 1, wherein the plurality of cancer classes is at least two or more cancer classes selected from the group consisting of non-cancer, bladder cancer, brain cancer, breast cancer, colorectal cancer, endometrial cancer, esophageal cancer, head/neck cancer, kidney cancer, liver cancer, hematological cancer, lung cancer, a lymphoma, leukemia, a melanoma, a lymphoma, ovarian cancer, pancreatic cancer, prostate cancer, rectal cancer, renal cancer, thyroid cancer and uterine cancer. ### Claim 5 (depends on 1) The method of claim 1, wherein: the biological sample or methylation biological sample obtained from the respective reference subject is a whole blood sample from the respective reference subject, and the nucleic acids in the biological sample or methylation biological sample obtained from the respective reference subject are genomic DNA. ### Claim 6 (depends on 1) The method of claim 1, wherein: the biological sample or methylation biological sample obtained from the respective reference subject is a plasma sample from the respective reference subject. ### Claim 7 (depends on 1) The method of claim 1, wherein the dimensionality reduction method further comprises application of principal component analysis using the cancer class and the sequencing construct of each reference subject in the first plurality of reference subjects thereby identifying the feature set comprising a plurality of principal components. ### Claim 8 (depends on 1) The method of claim 1, further comprising regularizing the first plurality of reference subjects set using the cancer class and the sequencing construct of each reference subject in the first plurality of reference subjects. ### Claim 9 (depends on 1) The method of claim 1, further comprising scaling the first bin count for each respective bin in the plurality of bins for each respective reference subject in the first plurality of reference subjects by: taking a log transformation of the respective first bin count thereby forming a log transformed first bin count for the respective bin, subtracting a mean value of the respective log transformed first bin count across the first plurality of reference subjects from the log transformed first bin count of the respective bin thereby forming a first normalized first bin count for the respective bin; and dividing, subsequent to the subtracting, the respective first normalized first bin count for the respective bin by a standard deviation of the first normalized bin first count across the first plurality of reference subjects thereby scaling the first bin count for each respective bin in the plurality of bins for each respective reference subject in the first plurality of reference subjects. ### Claim 10 (depends on 1) The method of claim 1, further comprising estimating a performance of the first trained classifier as an average performance of the trained component classifier formed in each respective iteration in the plurality of training iterations. ### Claim 11 (depends on 1) The method of claim 1, wherein: the sequencing construct for the respective reference subject further includes a second bin count for each respective bin in the plurality of bins, each respective second bin count representative of a number of nucleic acid fragments that are in a second size range that were measured from nucleic acids in the biological sample obtained from the respective reference subject that maps onto the different and non-overlapping portion of the reference genome, each respective first bin count representative of a number of nucleic acid fragments that are in a first size range that were measured from nucleic acids in the biological sample obtained from the respective reference subject that maps onto the different and non-overlapping portion of the reference genome, the collectively subjecting step further provides the second bin count of each bin in each respective plurality of bins across the first plurality of reference subjects to the dimensionality reduction method thereby obtaining the feature set, and the first size range is different than the second size range. ### Claim 12 (depends on 1) The method of claim 1, wherein: the sequencing construct for the respective reference subject includes a respective set of bin counts for each respective bin in the plurality of bins, wherein the respective set of bin counts includes the first bin count, and wherein each respective bin count in the respective set of bin counts is representative of a number of nucleic acid fragments that are in a size range corresponding to the respective bin count that were measured from nucleic acids in the biological sample obtained from the respective reference subject that maps onto the different and non-overlapping portion of the reference genome, and the collectively subjecting step provides the respective set of bin counts of each bin in the plurality of bins across the first plurality of reference subjects to the dimensionality reduction method thereby obtaining the feature set, and wherein the respective set of bins includes at least three different bin counts, and wherein each bin count in the respective set of bin counts corresponds to a different size range. ### Claim 13 A method of classifying a test subject of a given species to a cancer class in a plurality of cancer classes using a computer system comprising one or more processors, the method comprising: using a trained first classifier to classify the test subject to a cancer class in the plurality of cancer classes using nucleic acid fragments in a biological sample obtained from the test subject, wherein the trained first classifier having been trained via: obtaining, by the computer system and for each respective reference subject in a first plurality of reference subjects, (i) a cancer class of the respective reference subject and (ii) a sequencing construct for the respective reference subject that includes a first bin count for each respective bin in a plurality of bins that collectively represent all or a portion of a reference genome of the species, wherein each respective first bin count representative of a number of nucleic acid fragments measured from nucleic acids in a biological sample obtained from the respective reference subject that maps onto a different and non-overlapping portion of the reference genome of the species, wherein, for each respective cancer class in the plurality of cancer classes, the first plurality of reference subjects includes at least one reference subject that has the respective cancer class; improving a computational efficiency of the computer system by collectively subjecting, by the computer system, the first bin count of each bin in the plurality of bins for each reference subject in the first plurality of reference subjects to a dimensionality reduction method thereby obtaining a feature set, wherein the feature set consists of a number of features that is fewer than the number of bins in the plurality of bins; resampling, using the computer system, the feature set a plurality of times, wherein the resampling comprises forming, for each respective training iteration in a plurality of training iterations, a trained component classifier via: omitting, from the feature set, a subset of values for features in the feature set for the first plurality of reference subjects; and forming the trained component classifier by inputting, in conjunction with the cancer class of respective reference subjects in the first plurality of reference subjects as ground truth, remaining values for the features in the feature set as collective input to a respective untrained component classifier; and constructing, as a result of the resampling and by collectively leveraging output generated by the trained component classifier formed in each of the plurality of training iterations, a trained first classifier having an improved cancer class recognition ability over the untrained first classifier. ### Claim 14 (depends on 13) The method of claim 13, wherein: the using step combines a first call on cancer class made by the trained first classifier for the test subject with a second call made by a trained second classifier, input to the second trained classifier for the second call comprises a methylation pattern measured in a methylation biological sample obtained from the test subject, and the trained second classifier having been trained using a respective methylation pattern measured in a respective reference methylation biological sample from each reference subject in a second plurality of reference subjects. ### Claim 15 (depends on 14) The method of claim 14, wherein the test subject is deemed to have a first cancer class in the plurality of cancer classes when both the first call and the second call identify the test subject as having the same cancer class in the plurality of cancer classes. ### Claim 16 (depends on 14) The method of claim 14, wherein the test subject is deemed to have a first cancer class in the plurality of cancer classes when (i) the trained first classifier calls the first cancer class with a higher probability than all other cancer classes in the plurality of cancer classes and (ii) the second call identifies the test subject as having the first cancer class. ### Claim 17 (depends on 14) The method of claim 14, wherein the test subject is deemed to have a first cancer class in the plurality of cancer classes when (i) the first trained classifier calls the first cancer class with a call that is among the top two cancer classes in the plurality of cancer classes in terms of probability and (ii) the second call identifies the test subject as having the first cancer class. ### Claim 18 (depends on 14) The method of claim 14, wherein the test subject is deemed to have a first cancer class in the plurality of cancer classes when (i) the first trained classifier calls the first cancer class with a higher probability than all other cancer classes in the plurality of cancer classes and (ii) the second trained classifier calls the first cancer class with a higher probability than all other cancer classes in the plurality of cancer classes. ### Claim 19 (depends on 14) The method of claim 14, wherein the test subject is deemed to have a first cancer class in the plurality of cancer classes when (i) the first trained classifier calls the first cancer class with a call that is among the top two cancer classes in the plurality of cancer classes in terms of probability and (ii) the second trained classifier calls the first cancer class with a call that is among the top two cancer classes in the plurality of cancer classes in terms of probability. ### Claim 20 (depends on 14) The method of claim 14, wherein: the biological sample obtained from the test subject is a plasma sample from the test subject, and the biological sample or the methylation biological sample obtained from the respective reference subject during the training was a plasma sample from the respective reference subject. ### Claim 21 (depends on 14) The method of claim 14, wherein: the classifying uses the trained first classifier to classify the test subject to the cancer class in the plurality of cancer classes and an aggressiveness of the cancer class using (i) nucleic acid fragments of cell-free nucleic acids in the biological sample or the methylation biological sample obtained from the test subject and (ii) an indication of whether a predetermined genetic marker is absent or present in the test subject, and the trained first classifier having been further trained such that: the obtaining step further comprises, for each respective reference subject in the first plurality of reference subjects, an indication of whether the predetermined genetic marker is absent or present in the respective reference subject, and the improving step further uses the indication of whether the predetermined genetic marker is absent or present in each respective reference subject in the first plurality of reference subjects as ground truth. ### Claim 22 (depends on 13) The method of claim 13, wherein: the trained first classifier is a multinomial classifier that provides a plurality of likelihoods responsive to the nucleic acid fragments obtained from cell-free nucleic acids from the test subject, wherein each respective likelihood in the plurality of likelihoods is a likelihood that the test subject has a corresponding cancer class in the plurality of cancer classes. ### Claim 23 (depends on 13) The method of claim 13, further comprising: identifying a recommended treatment to the test subject based upon the cancer class of the test subject determined by the first trained classifier. ### Claim 24 (depends on 13) The method of claim 13, further comprising: identifying a recommended first treatment to the test subject when the first trained classifier determines that the test subject has a first cancer, and identifying a recommended second treatment to the test subject when the first trained classifier determines that the test subject has a second cancer, wherein the first cancer and the second cancer are different cancers. ### Claim 25 (depends on 24) The method of claim 24, wherein the first cancer and the second cancer are each independently selected from the group consisting of bladder cancer, brain cancer, breast cancer, colorectal cancer, endometrial cancer, esophageal cancer, head/neck cancer, kidney cancer, liver cancer, hematological cancer, lung cancer, a lymphoma, leukemia, a melanoma, a lymphoma, ovarian cancer, pancreatic cancer, prostate cancer, rectal cancer, renal cancer, thyroid cancer and uterine cancer. ### Claim 26 A computer system comprising: one or more processors; and a non-transitory computer-readable medium including one or more sequences of instructions that, when executed by the one or more processors, cause the one or more processors to perform a method of training an untrained first classifier to classify a test subject of a given species to a cancer class in a plurality of cancer classes, the method comprising: obtaining, by the computer system and for each respective reference subject in a first plurality of reference subjects, (i) a cancer class of the respective reference subject and (ii) a sequencing construct for the respective reference subject that includes a first bin count for each respective bin in a plurality of bins that collectively represent all or a portion of a reference genome of the species, wherein each respective first bin count representative of a number of nucleic acid fragments measured from nucleic acids in a biological sample obtained from the respective reference subject that maps onto a different and non-overlapping portion of the reference genome of the species, wherein, for each respective cancer class in the plurality of cancer classes, the first plurality of reference subjects includes at least one reference subject that has the respective cancer class; improving a computational efficiency of the computer system by collectively subjecting, by the computer system, the first bin count of each bin in the plurality of bins for each reference subject in the first plurality of reference subjects to a dimensionality reduction method thereby obtaining a feature set, wherein the feature set consists of a number of features that is fewer than the number of bins in the plurality of bins; resampling, using the computer system, the feature set a plurality of times, wherein the resampling comprises forming, for each respective training iteration in a plurality of training iterations, a trained component classifier via: omitting, from the feature set, a subset of values for features in the feature set for the first plurality of reference subjects; and forming the trained component classifier by inputting, in conjunction with the cancer class of respective reference subjects in the first plurality of reference subjects as ground truth, remaining values for the features in the feature set as collective input to a respective untrained component classifier; and constructing, as a result of the resampling and by collectively leveraging output generated by the trained component classifier formed in each of the plurality of training iterations, a trained first classifier having an improved cancer class recognition ability over the untrained first classifier. ### Claim 27 A computer system, comprising: one or more processors; and a non-transitory computer-readable medium including one or more sequences of instructions that, when executed by the one or more processors, cause the one or more processors to perform a method of classifying a test subject of a given species to a cancer class in a plurality of cancer classes, the method comprising: using a trained first classifier to classify the test subject to a cancer class in the plurality of cancer classes using nucleic acid fragments in a biological sample obtained from the test subject, the trained first classifier having been trained via: obtaining, by the computer system and for each respective reference subject in a first plurality of reference subjects, (i) a cancer class of the respective reference subject and (ii) a sequencing construct for the respective reference subject that includes a first bin count for each respective bin in a plurality of bins that collectively represent all or a portion of a reference genome of the species, wherein each respective first bin count representative of a number of nucleic acid fragments measured from nucleic acids in a biological sample obtained from the respective reference subject that maps onto a different and non-overlapping portion of the reference genome of the species, wherein, for each respective cancer class in the plurality of cancer classes, the first plurality of reference subjects includes at least one reference subject that has the respective cancer class; improving a computational efficiency of the computer system by collectively subjecting, by the computer system, the first bin count of each bin in the plurality of bins for each reference subject in the first plurality of reference subjects to a dimensionality reduction method thereby obtaining a feature set, wherein the feature set consists of a number of features that is fewer than the number of bins in the plurality of bins; resampling, using the computer system, the feature set a plurality of times, wherein the resampling comprises forming, for each respective training iteration in a plurality of training iterations, a trained component classifier via: omitting, from the feature set, a subset of values for features in the feature set for the first plurality of reference subjects; and forming the trained component classifier by inputting, in conjunction with the cancer class of respective reference subjects in the first plurality of reference subjects as ground truth, remaining values for the features in the feature set as collective input to a respective untrained component classifier; and constructing, as a result of the resampling and by collectively leveraging output generated by the trained component classifier formed in each of the plurality of training iterations, a trained first classifier having an improved cancer class recognition ability over the untrained first classifier.