Machine Learning for Somatic Single Nucleotide Variant Detection in Cell-free Tumor Nucleic acid Sequencing Applications

A system for detecting a presence or an absence of a tumor-derived somatic single-nucleotide variation (SNV) in a cell-free deoxyribonucleic acid (DNA) sample of a human subject having a cancer, comprising:

• computer memory comprising a trained machine learning classifier, wherein the trained machine learning classifier is trained by:
• (i) generating training data with class labels, wherein the training data comprises a plurality of mapped sequence reads generated from a plurality of training cell-free DNA samples comprising tumor-derived cell-free DNA molecules from human subjects having cancer;
• (ii) in computer memory, constructing the machine learning classifier configured to receive input features extracted from the plurality of mapped sequence reads of the plurality of training cell-free DNA samples and generate outputs for each of adenine (A), cytosine (C), guanine (G), and thymine (T) base calls based on the input features;
• and
• (iii) training the machine learning classifier with the training data;
• and
• one or more computer processors that are individually or collectively programmed to:
• (a) sequence a plurality of tumor-derived cell-free DNA molecules from a cell-free DNA sample obtained from the human subject having the cancer, to produce a set of tumor-derived cell-free DNA sequence reads;
• (b) map a plurality of the set of tumor-derived cell-free DNA sequence reads to a reference sequence to produce a set of mapped sequence reads;
• (c) apply the trained machine learning classifier to the set of mapped sequence reads to generate a plurality of base calls at a genomic base position among the set of mapped sequence reads that corresponds to the tumor-derived somatic SNV;
• (d) determine a number or a frequency of mapped sequence reads from among the set of mapped sequence reads having one or more variant base calls from among the plurality of base calls at the genomic base position that differ from the reference sequence;
• and
• (e) detect the presence or the absence of the tumor-derived somatic SNV in the cell-free DNA sample of the human subject having the cancer, based at least in part on the number or the frequency of the mapped sequencing reads determined in (d).

A non-transitory computer-readable medium comprising machine executable code that, upon execution by one or more computer processors, implements a method for detecting a presence or an absence of a tumor-derived somatic single-nucleotide variation (SNV) in a cell-free deoxyribonucleic acid (DNA) sample of a human subject having a cancer, the method comprising:

• (a) sequencing a plurality of tumor-derived cell-free DNA molecules from a cell-free DNA sample obtained from the human subject having the cancer, to produce a set of tumor-derived cell-free DNA sequence reads;
• (b) mapping a plurality of the set of tumor-derived cell-free DNA sequence reads to a reference sequence to produce a set of mapped sequence reads;
• (c) applying a trained machine learning classifier to the set of mapped sequence reads to generate a plurality of base calls at a genomic base position among the set of mapped sequence reads that corresponds to the tumor-derived somatic SNV, wherein the trained machine learning classifier is trained by:
• (i) generating training data with class labels, wherein the training data comprises a plurality of mapped sequence reads generated from a plurality of training cell-free DNA samples comprising tumor-derived cell-free DNA molecules from human subjects having cancer;
• (ii) constructing the machine learning classifier configured to receive input features extracted from the plurality of mapped sequence reads of the plurality of training cell-free DNA samples and generate outputs for each of adenine (A), cytosine (C), guanine (G), and thymine (T) base calls based on the input features;
• and
• (iii) training the machine learning classifier with the training data;
• (d) determining a number or a frequency of mapped sequence reads from among the set of mapped sequence reads having one or more variant base calls from among the plurality of base calls at the genomic base position that differ from the reference sequence;
• and
• (e) detecting the presence or the absence of the tumor-derived somatic SNV in the cell-free DNA sample of the human subject having the cancer, based at least in part on the number or the frequency of the mapped sequencing reads determined in (d).