Computational modeling of loss of function based on allelic frequency
The disclosure relates to computer technology for precision diagnosis of various states of genetic material such as a gene sequenced from cell-free DNA in a sample. The state may include a somatic homozygous deletion, a somatic heterozygous deletion, a copy number variation, or other states. A computer system may generate competing probabilistic models that each output a probability that the genetic material is in a certain state. Each model may be trained on a training sample set to output a probability that the genetic material is in a respective state. In some embodiments, the computer system may use various probabilistic distributions to generate the models. For example, the computer system may use a beta-binomial distribution, a binomial distribution, a normal (also referred to as “Gaussian”) distribution, or other type of probabilistic modeling techniques.
1 . A system to distinguish between a somatic homozygous deletion and a somatic heterozygous deletion of a gene in a sample of cell-free nucleic acid molecules that does not exhibit germline deletion of the gene, the system comprising:
a sequencing system configured to receive and process the sample of cell-free nucleic acid molecules collected from a subject, the sequencing system comprising:
a sequencing pipeline having one or more sequencing devices for associating cell-free nucleic acid molecules of the sample collected from the subject with sequence reads;
a processor programmed to:
receive, via a sequence analysis pipeline, sequence reads from the sequencing system;
receive, via a pre-processing subsystem, data from the sequence analysis pipeline;
generate, via the pre-processing subsystem, training datasets;
store the training datasets in a training datastore;
based on the training datasets, implement a classifier to generate, via a first probabilistic distribution, a first model of allelic counts based on one or more germline heterozygous single nucleotide polymorphism (SNP) positions associated with the gene in the sample, the first model representing the somatic homozygous deletion;
based on the training datasets, implement the classifier to generate, via a second probabilistic distribution, a second model of allelic counts in the sample based on the one or more germline heterozygous SNP positions, the second model representing the somatic heterozygous deletion;
implement the classifier to compare a first output of the first model and a second output of the second model;
implement the classifier to generate a prediction that the somatic homozygous deletion for the gene exists in the sample of cell-free nucleic acid molecules based on the comparison, thereby improving the accuracy of detecting the somatic homozygous deletion in the sample of cell-free nucleic acid molecules, wherein detecting the existence of the somatic homozygous deletion in the sample of cell-free nucleic acid molecules indicates the subject as one who has homologous repair deficiency (HRD) and may benefit from an anti-cancer therapy; and
generating a precision treatment plan tailored to the subject based on the prediction that the somatic homozygous deletion for the genes exists in the sample of cell-free nucleic acid molecules, wherein the precision treatment plan comprises administering a poly ADP ribose polymerase (PARP) inhibitor to the subject.
2 . The system of claim 1 , wherein the first model represents a first probability that the sample includes the somatic homozygous deletion and the second model represents a second probability that the sample includes the somatic heterozygous deletion.
3 . The system of claim 1 , wherein the first probabilistic distribution is a same type of probabilistic distribution as the second probabilistic distribution.
4 . The system of claim 1 , wherein to generate the first model, the processor is programmed to determine one or more parameters for input to the first probabilistic distribution.
5 . The system of claim 4 , wherein the first probabilistic distribution comprises a type of probabilistic distribution comprising one of: a beta-binomial distribution, a binomial distribution, or a normal distribution.
6 . The system of claim 4 , wherein to generate the first model of allelic counts, the processor is further programmed to:
determine, for input to the first probabilistic distribution, a prevalence of heterozygosity of the one or more germline SNPs in a training set of samples.
7 . The system of claim 6 , wherein the training set of samples comprises a plurality of samples in which tumor is not detected (TND).
8 . The system of claim 6 , wherein to generate the first model of allelic counts, the processor is further programmed to:
determine, for input to the first probabilistic distribution, a standard deviation of a minor allele frequency (MAF) associated with the one or more germline SNPs in the training set of samples.
9 . The system of claim 8 , wherein to generate the first model, the processor is further programmed to:
determine, for input to the first probabilistic distribution, a number of molecules in the sample that supports a mutant allele.
10 . The system of claim 9 , wherein to generate the first model, the processor is further programmed to:
determine, for input to the first probabilistic distribution, a total number of molecules in the sample.
11 . The system of claim 10 , wherein to generate the first model, the processor is further programmed to:
calculate a first likelihood of the allelic counts of the one or more germline SNP positions in the sample assuming somatic homozygous deletion based on molecule coverage associated with the somatic homozygous deletion.
12 . The system of claim 11 , wherein to generate the second model, the processor is further programmed to:
calculate a second likelihood of the allelic counts of the one or more germline SNP positions in the sample assuming somatic heterozygous deletion based on molecule coverage associated with the somatic heterozygous deletion.
13 . The system of claim 4 , wherein to generate the second model, the processor is further programmed to:
determine, for input to the second probabilistic distribution for the second model, a mean of tumor fraction estimated from the sample.
14 . The system of claim 13 , wherein the tumor fraction is estimated based on sequence coverage information.
15 . The system of claim 13 , wherein to generate the second model, the processor is further programmed to:
determine, for input to the second probabilistic distribution for the second model, a standard deviation of tumor fraction estimated from the sample.
16 . The system of claim 1 , wherein the processor is further programmed to:
access a plurality of samples;
identify a set of samples from among the plurality of samples that include a germline deletion;
filter out the set of samples from the plurality of samples; and
identify, from among the filtered plurality of samples, a presence of the somatic homozygous deletion or the somatic heterozygous deletion.
17 . The system of claim 1 , wherein the first output comprises a first probability of a presence of the somatic homozygous deletion and the second output comprises a second probability of a presence of the somatic heterozygous deletion.
18 . The system of claim 1 , wherein to compare the first output of the first model and the second output of the second model, the processor is further programmed to:
execute a log likelihood function based on the first output and the second output.
19 . The system of claim 1 , wherein the gene comprises one of: BRCA1, BRCA2, and ATM.