Detection of human leukocyte antigen loss of heterozygosity

1. A computer-implemented method of detecting loss of heterozygosity (LOH) of a human leukocyte antigen (HLA) gene in a subject, the method comprising:

obtaining HLA coverage feature metrics of a biological sample;

providing one or more of the HLA coverage feature metrics to a three-class HLA loss of heterozygosity (LOH) modeling process trained to classify the biological sample as corresponding to one of three LOH classes, no LOH, partial LOH, or clonal LOH and determining the LOH class of the sample and determining, using the three-class HLA LOH modeling process, the LOH class for the HLA gene; and

generating and storing a report of the determined LOH class for the HLA gene.

2. The method of claim 1 , wherein the three-class HLA LOH modeling process is a sequential two stage modeling process having a first LOH classifier model stage and a second LOH classifier model stage, wherein providing the one or more of the HLA coverage feature metrics to the three-class HLA LOH modeling process comprises:

providing at least one of the HLA coverage feature metrics to the first LOH classifier model and determining either no LOH or a LOH for the sample; and

in response to determining LOH for the samples, providing at least one of the HLA coverage feature metrics to the second LOH classifier model stage and determining the LOH class as either partial LOH or clonal LOH for the sample.

3. The method of claim 2 , wherein the one or more of the HLA coverage feature metrics comprises:

read depth of coverage of candidate HLA allele of the HLA gene;

a ratio of a B allele frequency (BAF) of a stable allele in a tumor sample of the biological sample to the BAF of the stable allele in a normal sample of the biological sample;

a difference between a log ratio (log R) of coverage for the stable allele between the tumor sample and the normal sample and a log R of coverage of a lost HLA allele of the HLA gene between the tumor sample and the normal sample;

tumor purity;

a ratio of a BAF of the lost allele in the tumor sample to the BAF of the lost allele in the normal sample; and

a quotient of the observed log R difference minus the expected log R difference divided by the expected log R difference based on tumor purity.

4. The method of claim 3 , wherein the observed log R difference is the difference between the log R of coverage of the stable allele and the log R of coverage of the lost allele.

5. The method of claim 4 , wherein the observed log R difference is an average of log(coverage in tumor/coverage in normal), calculated for at least one nucleotide position in an HLA gene.

6. The method of claim 5 , wherein the log(coverage in tumor/coverage in normal) is calculated for nucleotide positions having a coverage of at least 40 sequence reads.

7. The method of claim 4 , the observed log R difference is an average of log(coverage in tumor/coverage in normal*match ratio), calculated for at least one nucleotide position in an HLA gene, wherein the match ratio is the ratio of the number of HLA reads in the normal to number of HLA reads in the tumor or the ratio of the number of unique reads in the normal sample to the number of unique reads in the tumor sample.

8. The method of claim 7 , wherein log(coverage in tumor/coverage in normal * match ratio) is calculated for nucleotide positions having a coverage of at least 40 sequence reads.

9. The method of claim 3 , wherein the observed log R difference is the cumulative area between the log R line associated with a first allele and the log R line associated with a second allele.

10. The method of claim 3 , wherein the expected log R difference is the log 2(1-tumor purity) and tumor purity is a value between 0 and 1.

11. The method of claim 3 , which further comprises:

for each gene, calculating a ratio of a BAF of a first allele in the tumor sample to the BAF of the first allele in the normal sample and calculating a ratio of a BAF of a second allele in the tumor sample to the BAF of the second allele in the normal sample; and

comparing each ratio and selecting the allele associated with the lowest ratio as the allele that is more likely to be lost, before running the modeling process.

12. The method of claim 1 , wherein obtaining HLA coverage feature metrics of the biological sample comprises:

receiving next generation sequencing data generated from the biological sample of the subject;

aligning the next generation sequencing data against a reference genome to determine a mapped reads dataset and an unmapped reads dataset;

providing at least the unmapped reads dataset to an HLA typing process to identify at least one candidate HLA allele for the HLA gene;

identifying a HLA sequence associated with each identified candidate HLA allele;

creating a HLA reference genome using each identified HLA sequence;

aligning the next generation sequencing data against the HLA reference genome and adjusting the HLA reference genome to account for a variant identified during the aligning; and

aligning the next generation sequencing data against the adjusted HLA reference genome and, in response, determining the HLA coverage feature metrics associated with one or more identified candidate HLA alleles.

13. The method of claim 1 , wherein obtaining HLA coverage feature metrics of the biological sample comprises:

receiving normal next generation sequencing data generated from a buffy coat preparation of a blood sample of the subject;

aligning the next generation sequencing data against a reference genome to determine a normal mapped reads dataset and a normal unmapped reads dataset;

receiving tumor next generation sequencing data generated from a tumor specimen of the subject;

providing at least a portion of the normal unmapped reads dataset to an HLA typing process to identify at least one candidate HLA allele for the HLA gene;

identifying a HLA sequence associated with each identified candidate HLA allele;

creating a HLA reference genome using each identified HLA sequence;

aligning the normal next generation sequencing dataset against the HLA reference genome and adjusting the HLA reference genome to account for a variant identified during the aligning; and

aligning the normal next generation sequencing dataset against the adjusted HLA reference genome and aligning the tumor next generation sequencing dataset against the adjusted HLA reference genome to determine the HLA coverage feature metrics associated with the identified candidate HLA alleles.

14. The method of claim 13 , wherein determining the LOH class for the HLA gene comprises applying a logistic regression model to the obtained HLA coverage feature metrics.

15. The method of claim 14 , wherein the one or more of the HLA coverage feature metrics comprises:

read depth of coverage of a candidate allele of the HLA gene;

a ratio of a B allele frequency (BAF) of a stable allele in a tumor sample of the biological sample to the BAF of the stable allele in a normal sample of the biological sample;

a difference between a log ratio (log R) of coverage for the stable allele between the tumor sample and the normal sample and a log R of coverage of a lost HLA allele of the HLA gene between the tumor sample and the normal sample;

tumor purity;

a ratio of a BAF of the lost allele in the tumor sample to the BAF of the lost allele in the normal sample; and

a quotient of the observed log R difference minus the expected log R difference divided by the expected log R difference based on tumor purity.

16. The method of claim 15 , wherein the next generation sequencing data is generated using short read sequencing.

17. A method for determining loss of heterozygosity for the HLA-A, HLA-B, and HLA-C genes, or for the HLA-E, HLA-F, and HLA-G genes, or for the DRA, DRB1, DQA1, DQB1, DPA1, and DPB1 genes using, for each gene, the method of claim 15 .

18. The method of claim 15 , wherein at least a portion of the reads data comprises forward reads from paired-end reads.

19. The method of claim 15 , wherein the HLA typing process applies an Optitype HLA typing algorithm or a Kourami HLA typing algorithm.

20. The method of claim 15 , wherein the HLA reference genome further comprises at least one HLA pseudogene sequence.

21. The method of claim 15 , wherein providing at least a portion of the normal unmapped reads dataset to the HLA typing process to identify at least one candidate HLA allele for the HLA gene comprises providing at least a portion of the normal unmapped reads dataset and a portion of the normal mapped reads dataset to the HLA typing process.

22. The method of claim 15 , wherein aligning the tumor next generation sequencing dataset against the adjusted HLA reference genome to determine the HLA coverage feature metrics comprises filtering the tumor next generation sequencing dataset.

23. The method of claim 22 , wherein filtering the tumor next generation sequencing dataset comprises removing reads that are not properly aligned, removing duplicate reads, and/or removing a read based on an edit distance associated with the read.

24. The method of claim 15 , wherein the tumor specimen is a solid tumor specimen.

25. The method of claim 15 , wherein the tumor specimen is a cell free DNA (cfDNA) specimen.

26. The method of claim 15 , wherein the tumor specimen is a lung tumor specimen, a metastatic specimen, a colorectal tumor specimen, or a pancreatic tumor specimen.

27. The method of claim 15 , wherein the method is implemented on one or more microservices.

28. The method of claim 15 , wherein the method further comprises:

for the biological sample containing cancer, when it is determined that the biological sample has an LOH class of no LOH in the HLA gene, treating the cancer by administering a checkpoint inhibitor therapy to the subject.

29. The method of claim 28 , wherein the checkpoint inhibitor therapy is selected from the group consisting of an anti-CTLA-4 therapy, an anti-PD-1 therapy, and an anti-PD-L1 therapy.

30. The method of claim 1 , wherein the biological sample is selected from the group consisting of a tumor specimen and a buffy coat preparation.