Dynamically selecting sequencing subregions for cancer classification

A method for classifying a variant in a cell free nucleic acid sample, the method comprising:

• training, using a training population, a machine-learned model as a cancer classifier to identify test sequences with candidate variants that are indicative of cancer and identify cancer in test samples using indicative test sequences, wherein the training population comprises one or more sequencing regions selected from an aggregated cohort of sequencing regions selected for indicating cancer presence;
• accessing a plurality of test sequences from a test sample with each test sequence of the plurality of test sequences comprising one or more sequencing regions of a plurality of sequencing regions, the plurality of sequencing regions, in aggregate, forming an aggregate sequencing region;
• selecting one or more indicative regions from the aggregate sequencing region formed from the plurality of test sequences to generate a sample population, the selected one or more indicative regions corresponding to a cancer presence in the test sample, wherein the sample population is smaller than the aggregate sequencing region;
• and
• applying the machine-learned model to the generated sample population to identify the cancer presence in the test sample, the machine-learned model:
• identifying a candidate variant in an indicative region of the one or more selected indicative regions for the sample population;
• comparing a test feature set of the candidate variant to a cohort feature set of a matching sequencing subregion from the cohort population;
• and
• determining the candidate variant is indicative of the cancer in the test sample based on the compared test feature set and cohort feature set.

A system comprising:

• one or more processors;
• and
• a non-transitory computer readable storage medium storing computer program instructions for classifying a variant in a cell free nucleic acid sample, the method comprising:
• train, using a training population, a machine-learned model as a cancer classifier to identify test sequences with candidate variants that are indicative of cancer and identify cancer in test samples using indicative test sequences, wherein the training population comprises one or more sequencing regions selected from an aggregated cohort of sequencing regions selected for indicating cancer presence;
• access a plurality of test sequences from a test sample with each test sequence of the plurality of test sequences comprising one or more sequencing regions of a plurality of sequencing regions, the plurality of sequencing regions, in aggregate, forming an aggregate sequencing region;
• select one or more indicative regions from the aggregate sequencing region formed from the plurality of test sequences to generate a sample population, the selected one or more indicative regions corresponding to a cancer presence in the test sample, wherein the sample population is smaller than the aggregate sequencing region;
• and
• apply the machine-learned model to the generated sample population to identify the cancer presence in the test sample, the machine-learned model:
• identifying a candidate variant in an indicative region of the one or more selected indicative regions for the sample population;
• comparing a test feature set of the candidate variant to a cohort feature set of a matching sequencing subregion from the cohort population;
• and
• determining the candidate variant is indicative of the cancer in the test sample based on the compared test feature set and cohort feature set.

A non-transitory computer readable storage medium storing computer program instructions for classifying a variant in a cell free nucleic acid sample, the method comprising:

• train, using a training population, a machine-learned model as a cancer classifier to identify test sequences with candidate variants that are indicative of cancer and identify cancer in test samples using indicative test sequences, wherein the training population comprises one or more sequencing regions selected from an aggregated cohort of sequencing regions selected for indicating cancer presence;
• access a plurality of test sequences from a test sample with each test sequence of the plurality of test sequences comprising one or more sequencing regions of a plurality of sequencing regions, the plurality of sequencing regions, in aggregate, forming an aggregate sequencing region;
• select one or more indicative regions from the aggregate sequencing region formed from the plurality of test sequences to generate a sample population, the selected one or more indicative regions corresponding to a cancer presence in the test sample, wherein the sample population is smaller than the aggregate sequencing region;
• and
• apply the machine-learned model to the generated sample population to identify the cancer presence in the test sample, the machine-learned model:
• identifying a candidate variant in an indicative region of the one or more selected indicative regions for the sample population;
• comparing a test feature set of the candidate variant to a cohort feature set of a matching sequencing subregion from the cohort population;
• and
• determining the candidate variant is indicative of the cancer in the test sample based on the compared test feature set and cohort feature set.