Systems and methods for phasing tumor mutations

Abstract

This application relates in general to the analysis of mutations in tumors, and more specifically to a system and method for phasing mutations in tumors of a subject (e.g., cancer patients). An exemplary method for phasing mutations in a tumor of a subject includes: enumerating sets of unique mutation patterns observed in multiple sequence reads based on tumor DNA and / or RNA sequence reads; counting sets of unique patterns observed in sequence reads to calculate the amount of each unique mutation pattern and / or the amount of each combination of a unique mutation pattern and a group of transcripts; determining the probability of the mutation patterns; and inputting the amount of mutation patterns and the probability of mutation patterns into a statistical model to estimate at least one of the following: a set of haplotype probability sets in which each haplotype exists, a set of haplotype incidence sets, and a set of haplotype transcript incidence sets.

Description

[Technical Field]

[0001] Cross-reference of related applications This application claims priority to U.S. Provisional Application No. 63 / 502,387, filed May 15, 2023, the entire contents of which are incorporated herein by reference.

[0002] Technical field This application relates in general to the analysis of mutations in tumors, and more specifically to a system and method for phasing mutations in tumors of subjects (e.g., cancer patients). [Background technology]

[0003] background The emergence of cancer immunology and next-generation sequencing (NGS) and bioinformatics technologies has advanced the development of personalized neoantigen-specific immunotherapies. Personalized neoantigen-specific immunotherapies may include any treatment or immunotherapy designed and developed based on each patient's specific tumor mutations, for example, with the aim of inducing a high-affinity T-cell response against cancer cells. Specifically, while tumor cells share most of their DNA with healthy cells, tumor cells also possess unique mutations. Genetic mutations can lead to the expression of unique tumor antigens called neoantigens. Neoantigens are mutated proteins whose fragments are presented to T cells via the major histocompatibility complex (MHC), thereby potentially driving anti-tumor immunity. Therefore, neoantigens have emerged as promising targets for immunotherapies that seek to induce the immune system to specifically destroy tumor cells.

[0004] Short-read DNA (and / or RNA) sequencing can be used to identify candidate neoantigens. Specifically, DNA can be extracted and sequenced from tumor tissue of an individual patient. Furthermore, the DNA can also be extracted and sequenced from one or more matched normal tissue samples from the same patient, or from normal tissue samples taken from one or more comparable individuals. Both tumor and normal short-read sequencing data can be aligned to a reference genome. If several aligned reads from a tumor sample do not match the reference genome at a certain position, but most aligned reads from a normal sample match the reference genome at that same position, this may indicate a somatic tumor mutation at that position. Such somatic mutations can be detected by a somatic mutation caller, which may use some type of statistical approach. As part of somatic mutation calling, the “variant allele frequency” (VAF) of the mutant allele can be estimated. High VAFs typically correspond to more clonal mutations, and therefore the corresponding peptide or protein (e.g., neoantigen) may be a better immunotherapy target because it can target more cells within the tumor. If multiple aligned reads from a normal sample do not match the reference genome at a certain position, this may indicate a germline variant that is likely present in at least some cells in normal tissue and also in some cells in tumor tissue. These variants can be detected through a germline variant caller. RNA reads can be used similarly to identify somatic and germline variants. Regardless of the source of the called variant, somatic and germline variants can then be projected onto the reference transcriptome (e.g., as specified in a Gene Transfer Format (GTF) file) and translated in silico (e.g., using Ensembl variant effect predictors (VEPs) and known codon triplet-versus-amino acid correspondences) to predict tumor peptides. The variant sequence of the tumor polypeptide (resulting from somatic mutations) corresponds to the predicted candidate neoantigen.

[0005] Somatic and germline mutant calling determines only the location of somatic and germline mutants occurring as a whole in at least one copy of a gene in several sequenced cells. However, whether using bulk sequencing or single-cell sequencing (and more broadly than in the context of tumor sequencing), multiple copies of a given gene are typically present in a given sample / cell being sequenced, and mutant calling itself does not reveal which subsets of somatic and germline mutant alleles occurred simultaneously in any one of these gene copies. In other words, several copies of a gene may contain one subset of the mutant alleles being called, while other copies of that gene may contain different subsets of the mutant alleles, and the sequencing data from all copies of the gene are typically mixed together. The combination of mutant alleles that occur simultaneously is often described as a haplotype. Determining which mutant alleles occur simultaneously in at least one copy of a gene (or one copy of a subregion of a gene) is referred to as "phasing" the mutants. In relation to somatic mutant phasing, phasing mutants (also known as "mutation phasing") is necessary to accurately perform in silico translation and predict sequential candidate neoantigens. Furthermore, since most mutants close to a given somatic mutant may be germline mutants, these mutants may need to be phased together with germline mutants.

[0006] Mutant phasing (somatic or germline) can be achieved directly through specific complementary experiments, but such approaches are low-throughput techniques that are not practical in clinical settings. On the other hand, various computational methods can phase germline mutants (not somatic mutants) and identify haplotypes present in a given sample from short-read NGS (or microarray) data. Some approaches for germline mutant phasing utilize population-reference-based phasing, which leverages the fact that a common ancestor and limited recombination result in common and inherited haplotype blocks, assuming that the haplotype of any new patient is a mosaic of haplotype blocks from a reference panel of known haplotypes. Other phasing techniques can avoid the reference population assumption and be limited to "readback" computational phasing. Specifically, readback phasing can be useful for identifying haplotypes based on high-throughput sequence reads that overlap with two or more mutants.

[0007] Unfortunately, the vast majority of existing computational techniques rely on the assumption that neither reference-based nor feedback is retained in tumor sequencing data, and thus are only suitable for phasing germline variants in healthy tissue. Reference-based techniques that predict the inheritance of haplotype blocks are clearly inappropriate because somatic mutations (by definition) do not inherit. More generally, however, most methods for computational phasing assume that there are two or fewer unique haplotypes for a given gene, which is not true when dealing with bulk data where cells from multiple tumor cell subpopulations within the same patient can be sequenced. This is not necessarily the case with single-cell sequencing of tumor cells where some cells may have undergone copy number amplification and subsequent mutations in a given gene. A smaller proportion of existing methods can phase polyploid genomes (e.g., genomes containing three or more haplotypes), but these methods still often assume that each haplotype is equally represented in the genome of the sequenced sample, an assumption that is again weakened by the presence of tumor subpopulations and copy number variations. For similar reasons, these polyploid phasing methods are inappropriate for RNA sequence read data because haplotypes can be expressed at different rates (even if they are present in equal proportions in genomic DNA).

[0008] Most algorithms phase out germline mutations in diploid samples. For example, some algorithms involve a combination of graphs such as Hapcut2 and the "max-cut" bipartition of mutations and read-back phasing. As another example, some algorithms (e.g., Shapeit4) involve a combination of population reference and hidden Markov model (HMM) phasing, where each individual is assumed to be a mosaic of reference haplotypes. Methods for polyploid phasing are much fewer, and these are generally not optimal for phasing tumor sequencing data. For example, Haptree would not be suitable for use with bulk tumor sequencing data as it assumes an equal dosage of haplotypes (i.e., that haplotypes are present in equal proportions) when assigning sequence reads to each haplotype. Further, this assumption is inappropriate for modeling RNA sequencing data or for explaining copy number variations (CNVs). Alternatively, Hapcompass is purely graph-based and does not make such assumptions, but at the same time, in normal polyploid phasing tasks, it usually performs worse compared to new methods.

[0009] In addition to the polyploid nature and unequal haplotype / expression levels unique to tumor samples, phasing is particularly hampered by the limitations of short reads, even for normal samples. Most short reads are not long enough to cover all variants within a gene simultaneously, so it does not clearly show which haplotypes are present in the samples that are observed. Further, as part of clinical sequencing, instead of a single sequencing sample for which other phasing methods are intended, it is possible to obtain T-N-R triplets of sequencing data (i.e., tumor DNA-normal DNA-RNA sequencing data). Therefore, new systems and methods suitable for phasing mutations in tumors of cancer patients are desirable. SUMMARY OF THE INVENTION

[0010] Abstract Methods, systems, and programming for phasing (a broad term for phasing somatic and / or germline variants) identified in a target tumor are described herein. The broad idea is that a particular type of recognized DNA sequence read and / or RNA sequence read is more likely to be generated by several haplotypes (and, in the case of RNA sequence reads, several transcripts) than by others, and a haplotype is defined as a given combination of mutant alleles (e.g., single nucleotide variants (SNVs), indels, etc.) occurring simultaneously in at least one copy of a gene or gene subregion (here again, the term “variant” in this context may refer to germline variants and / or somatic variants). Thus, DNA and / or RNA sequence reads provide evidence of the presence of a particular haplotype and / or the incidence of a particular haplotype and / or (if RNA sequence read data is available) the expression level of a particular haplotype in a particular transcript.

[0011] In some embodiments, multiple sequence reads derived from tumor cells obtained from a subject can be accessed. These multiple sequence reads (e.g., single reads or paired-end reads) may include tumor DNA sequence reads and / or tumor RNA sequence reads. A set of unique mutation patterns (i.e., all possible unique combinations of observed mutant alleles that may occur in a given gene sequence or part thereof) can be enumerated based on a set of mutations (e.g., tumor-associated mutations) identified in the multiple sequence reads (a given mutant allele may or may not be observed in a given individual sequence read). The quantity for each unique mutation pattern can then be calculated by counting the number of times each unique mutation pattern in the set is observed in the multiple sequence reads. For each unique mutation pattern, and for each haplotype in the set of all candidate haplotypes (the set of candidate haplotypes is the set of all possible combinations of observable mutant alleles, regardless of the actual sequence read data (e.g., in the case of bi-allele mutations, a set containing 2^(num_mutations) haplotypes)), the probability that a hypothetical DNA sequence read from that haplotype exhibits a unique mutation pattern can be determined. Similarly, if RNA sequence reads are available, for each possible unique mutation pattern, each possible unique transcript group (described below), each candidate haplotype, and each transcript, the probability that a hypothetical RNA sequence read from a given haplotype and transcript exhibits a unique mutation pattern and simultaneously matches a given unique transcript group can be determined (i.e., each unique transcript group is one of the non-empty subsets of transcripts transcribed by the phasing gene, i.e., there are 2^(number of gene transcripts-1) transcript groups; a base in a sequence read matches a transcript group if that base aligns to a position in an exon that is expressed only by the transcripts in that transcript group and not by other transcripts; a sequence read matches the smallest transcript group to which its base matches).

[0012] Each haplotype-transcript combination can generate RNA sequence reads that potentially exhibit a given combination of mutation patterns and transcripts associated with a given gene. Thus, haplotype-transcript combinations correspond to the underlying genomic reality. Mutation patterns and transcripts correspond to noisy and / or ambiguous observations (of the form of DNA or RNA sequence reads) of the underlying reality. Each haplotype-transcript combination can result in multiple different combinations of mutation patterns and transcripts. Each combination of mutation patterns and transcripts may have arisen from multiple pairwise combinations of haplotypes and transcripts. Therefore, the correspondences between pairwise combinations of haplotypes and transcripts, and between pairwise combinations of mutation patterns and transcripts, are many-to-many relationships.

[0013] Next, the amount and probability of unique mutation patterns determined for DNA sequence reads, and / or the amount and probability of unique mutation pattern transcripts determined for RNA sequence reads, can be input into a statistical model to estimate at least one of the following: (i) a set of haplotype probabilities (i.e., the probability that each haplotype is present in the sample), (ii) a set of haplotype incidence rates (i.e., the incidence rate of each haplotype in the set of identified haplotypes, expressed, for example, as a percentage of the total number of DNA molecules in the sequenced sample that are each haplotype), and / or (iii) a set of haplotype transcript abundance rates (i.e., the incidence rate of each haplotype transcript combination in the set of identified haplotype transcript combinations, expressed, for example, as a percentage of the total number of RNA molecules in the sequenced sample that are each haplotype transcript combination). The disclosed method can be carried out using DNA sequence reads, RNA sequence reads, or a combination thereof if both DNA sequence reads and RNA sequence reads are available.

[0014] In some embodiments, a set of haplotype probabilities, haplotype incidence rates, and / or haplotype transcript incidence rates determined by a statistical model are used to identify a set of tumor-associated peptides or proteins (e.g., neoantigens) and / or select a subset of those tumor-associated peptides or proteins for use in the development of patient-specific therapies (e.g., anti-cancer therapies). For example, a set of haplotypes identified by sequence read data (e.g., based on a comparison of haplotype probabilities with a predetermined threshold) can be translated in silico to determine tumor-associated peptide or protein sequences (e.g., by parsing nucleotide sequence data into codons and then translating the codons into corresponding amino acid sequences). Haplotype incidence data output by the statistical model can be used to rank and / or select a subset of tumor-associated peptide or protein sequences for use in the development of patient-specific therapies (e.g., anti-cancer therapies). Using the incidence data of haplotype transcripts output by statistical models, the ranking and / or selection of subsets of tumor-associated peptide or protein sequences for use in the development of patient-specific therapies (e.g., anti-cancer therapies) can be further improved by determining, for example, that the transcript expression of one tumor-associated peptide or protein (and its associated haplotype) is far more abundant than that of the same transcript of a different tumor-associated peptide or protein (and its different associated haplotype).

[0015] Some embodiments of this disclosure include a system comprising one or more data processors. In some embodiments, the system includes a non-temporary computer-readable storage medium containing instructions that, when executed by one or more data processors, cause one or more data processors to perform steps of any of the methods disclosed herein.

[0016] Some embodiments of the present disclosure include computer program products tangibly embodied in a non-temporary machine-readable storage medium, which include instructions configured to cause one or more data processors to perform steps of any of the methods disclosed herein.

[0017] Some embodiments of the present disclosure include a vaccine comprising one or more peptides, a plurality of nucleic acids encoding one or more peptides, or a plurality of cells expressing one or more peptides, wherein the one or more peptides are selected from a set of peptides identified by performing a step of any of the methods disclosed herein.

[0018] Some embodiments of the present disclosure include a method for designing a vaccine comprising one or more peptides, a plurality of nucleic acids encoding one or more peptides, or a plurality of cells expressing one or more peptides, the method comprising identifying one or more peptides using the method described herein.

[0019] Some embodiments of the present disclosure include a method for producing a vaccine, comprising producing a vaccine comprising one or more peptides, a plurality of nucleic acids encoding one or more peptides, or a plurality of cells expressing one or more peptides, wherein the one or more peptides are selected from a set of peptides identified by performing a step of any of the methods disclosed herein.

[0020] Some embodiments of this disclosure include a pharmaceutical composition comprising one or more peptides selected from a set of peptides identified by performing steps of any of the methods disclosed herein.

[0021] A method for phasing mutations identified in a target tumor, comprising: accessing multiple sequence reads derived from tumor cells obtained from the target using one or more computing devices, wherein the sequence reads include tumor DNA sequence reads and / or tumor RNA sequence reads; enumerating sets of unique mutation patterns observed in the multiple sequence reads; counting the number of sequence reads representing each unique mutation pattern in the set of unique mutation patterns observed in the sequence reads to calculate the amount of each unique mutation pattern; and / or counting the number of sequence reads representing each combination of a unique mutation pattern from the set of unique mutation patterns and a set of transcripts from one or more transcripts associated with the gene to calculate the amount of each combination of unique mutation pattern and transcript; and for each unique mutation pattern, for each haplotype in the set of haplotypes, a hypothetical DNA sequence from the haplotype. A method is disclosed herein that includes determining the probability that a read exhibits a unique mutation pattern and / or the probability that a hypothetical RNA sequence read from a haplotype and transcript exhibits a unique mutation pattern and transcript group for each combination of haplotypes from a set of haplotypes, transcripts from a set of gene-associated transcripts, and transcript groups from a set of gene-associated transcript groups; inputting the amount of unique mutation patterns and / or the amount of unique mutation pattern transcript groups, as well as the probability of unique mutation patterns and / or the probability of unique mutation pattern transcript groups, into a statistical model to estimate at least one of the following: (i) a set of haplotype existence probabilities in which each haplotype exists, (ii) a set of haplotype occurrence rates, and (iii) a set of haplotype transcript occurrence rates; and outputting at least one of the following: (i) a set of haplotype existence probabilities, (ii) a set of haplotype occurrence rates, and (iii) a set of haplotype transcript occurrence rates.

[0022] In some embodiments, an estimate of at least one of a set of haplotype probability of presence, a set of haplotype incidence rates, and a set of haplotype transcript incidence rates includes (i) using a statistical model to sample for each haplotype in the set from at least one of the haplotype probability posterior distribution, the haplotype incidence posterior probability distribution, and the haplotype transcript incidence posterior probability distribution, and (ii) using the samples from each posterior probability distribution to calculate point estimates for the haplotype probability of presence, the haplotype incidence rate, and the haplotype transcript incidence rate for each haplotype in the set.

[0023] In some embodiments, the method further includes identifying a set of mutant peptide sequences using in silico translation of one or more haplotype sequences and / or haplotype transcripts based on a set of haplotype probability, a set of haplotype incidence rates, and / or haplotype transcript incidence rates. In some embodiments, one or more haplotype sequences and / or haplotype transcripts are associated with non-zero haplotype probability.

[0024] In some embodiments, the method further includes selecting one or more mutant peptide sequences from the set of mutant peptide sequences using one or more predetermined criteria, which include predetermined criteria applied to a set of haplotype incidence rates and / or a set of haplotype transcript incidence rates. In some embodiments, one or more predetermined criteria include predetermined haplotype incidence thresholds and / or predetermined haplotype transcript thresholds.

[0025] In some embodiments, the method further includes selecting one or more mutant peptide sequences from a set of mutant peptide sequences by ranking a set of peptide sequences based on a set of haplotype incidence rates and / or a set of haplotype transcript incidence rates.

[0026] In some embodiments, the method further includes using a machine learning model to generate predictions of the likelihood of presentation in major histocompatibility complexes (MHCs) for one or more of a set of mutant peptide sequences and / or predictions of immunogenicity for one or more of the set of mutant peptide sequences.

[0027] In some embodiments, accessing sequence reads further includes accessing multiple normal DNA sequence reads derived from healthy cells obtained from a subject. In some embodiments, counting the set of unique patterns observed in the multiple sequence reads further includes calculating the amount of each unique mutation pattern in the normal DNA sequence reads.

[0028] In some embodiments, counting sets of unique patterns observed in multiple sequence reads further includes calculating the amount of each unique mutation pattern in tumor DNA sequence reads and calculating the amount of each unique mutation and transcript group pattern in tumor RNA sequence reads.

[0029] In some embodiments, the probability that a hypothetical RNA sequence read from the haplotype and transcript exhibits a unique mutation pattern and transcript group is calculated from the haplotype transcript incidence, the conditional probability of accepting a combination of a unique mutation pattern and transcript group within the RNA sequence read, and the transcript length. In some embodiments, the calculation of the probability that a hypothetical RNA sequence read from the haplotype and transcript exhibits a unique mutation pattern and transcript group further includes considering the probability of misrecalling a tumor RNA sequence read that has a unique mutation pattern.

[0030] In some embodiments, the probability that a hypothetical DNA sequence read from that haplotype exhibits a unique mutation pattern is calculated from the haplotype incidence and the conditional probability of recognizing a unique mutation pattern in the DNA sequence read. In some embodiments, the calculation of the probability that a hypothetical DNA sequence read from that haplotype exhibits a unique mutation pattern further includes considering the probability of misrepresenting a tumor DNA sequence read as having a unique mutation pattern.

[0031] In some embodiments, the calculation of the probability that a hypothetical RNA sequence from its haplotype and transcripts exhibits a unique mutation pattern and transcript group further includes considering the probability that a given insert length in the RNA sequence read exhibits a unique mutation pattern and transcript group. In some embodiments, the calculation of the probability that a hypothetical DNA sequence from its haplotype exhibits a unique mutation pattern further includes considering the probability that a given insert length in the DNA sequence read exhibits a unique mutation pattern.

[0032] In some embodiments, the calculation of the probability that a hypothetical RNA sequence from the haplotype and transcripts exhibits a unique mutation pattern and transcript group further includes considering the probability of RNA sequencing error. In some embodiments, the calculation of the probability that a hypothetical DNA sequence from the haplotype and transcripts exhibits a unique mutation pattern and transcript group further includes considering the probability of DNA sequencing error.

[0033] In some embodiments, an estimate of at least one of a set of haplotype presence probabilities, a set of haplotype incidence rates, and a set of haplotype transcript incidence rates is obtained for each haplotype in the set of possible haplotypes by using a statistical model to sample the posterior probability distribution for haplotype presence to determine a point estimate for haplotype presence, sample the posterior probability distribution for haplotype incidence rates to determine a point estimate for haplotype incidence rates, and / or sample the posterior probability distribution for haplotype transcript incidence rates to determine a point estimate for haplotype transcript incidence rates. In some embodiments, the point estimates for each posterior probability distribution include a mean. In some embodiments, the posterior probability distributions for haplotype presence, haplotype incidence rates, and / or haplotype transcript incidence rates are further used to determine uncertainties associated with haplotype presence, haplotype incidence rates, and / or haplotype transcript incidence rates, respectively.

[0034] In some embodiments, accessing multiple sequence reads further includes accessing a set of germline variants and somatic variant calls derived from tumor cells obtained from the subject.

[0035] In some embodiments, the statistical model includes a probabilistic graphical model. In some embodiments, the statistical model includes a hierarchical Bayesian model. In some embodiments, the hierarchical Bayesian model includes a haplotype generation model, a DNA Dirichlet-multinomial model, and / or an RNA Dirichlet-multinomial model.

[0036] A system including one or more computing devices, Also disclosed herein is a system comprising one or more non-temporary computer-readable storage media containing instructions, and one or more processors coupled to one or more storage media, wherein one or more processors are configured to execute instructions for performing any of the methods described herein.

[0037] Also disclosed herein are non-temporary computer-readable media containing instructions that cause one or more processors of one or more computing devices to perform any of the methods described herein when executed by one or more processors of one or more computing devices.

[0038] A vaccine is disclosed herein, comprising one or more peptides, a plurality of nucleic acids encoding one or more peptides, or a plurality of cells expressing one or more peptides, wherein one or more peptides are selected from a set of peptides by performing any of the methods described herein.

[0039] A method for producing a vaccine is disclosed herein, comprising performing any of the methods described herein to select one or more peptides from a set of peptides, and producing a vaccine comprising at least one of the selected peptides; a plurality of nucleic acids encoding at least one of the selected peptides; or a plurality of cells expressing at least one of the selected peptides.

[0040] A pharmaceutical composition comprising one or more peptides selected from a set of peptides is disclosed herein, which is obtained by performing any of the methods described herein.

[0041] The terms and expressions used are for illustrative purposes only, not limitation, and in using such terms and expressions there is no intention to exclude equivalents or parts of the features shown and described, however it is acknowledged that various modifications are possible within the scope of the invention as described in the claims. Accordingly, although the claimed invention is specifically disclosed by embodiments and optional features, it should be understood that modifications and variations of the concepts disclosed herein can be used by those skilled in the art, and such modifications and variations are deemed to be within the scope of the invention as defined by the appended claims. [Brief explanation of the drawing]

[0042] [Figure 1A] This section illustrates various challenges in exemplary fading analysis. [Figure 1B] This section illustrates various challenges in exemplary fading analysis. [Figure 1C] This section illustrates various challenges in exemplary fading analysis.

[0043] [Figure 1D] This illustrates the challenges in exemplary readback phasing analysis of DNA reads.

[0044] [Figure 1E] This illustrates the challenges in exemplary readback phasing analysis of RNA sequence reads.

[0045] [Figure 2A] This specification illustrates exemplary processes for phasing mutations in a patient's tumor, according to several embodiments disclosed herein.

[0046] [Figure 2B] This specification discloses several embodiments of an exemplary tumor mutation phasing system for performing one or more methods for phasing mutations in a patient's tumor.

[0047] [Figure 3] This specification provides exemplary sequence read counters for enumerating and quantifying unique mutation patterns observed in RNA and / or DNA sequence reads according to several embodiments disclosed herein.

[0048] [Figure 4] This specification provides exemplary sequence read counters for enumerating and quantifying unique mutation patterns and transcript sets observed in RNA sequence reads according to several embodiments disclosed herein.

[0049] [Figure 5A] This specification presents exemplary enumerations and pattern estimators for determining the probability that hypothetical DNA sequence reads from a haplotype exhibit a unique mutation pattern, based on several embodiments disclosed herein. [Figure 5B] This specification presents exemplary enumerations and pattern estimators for determining the probability that hypothetical DNA sequence reads from a haplotype exhibit a unique mutation pattern, based on several embodiments disclosed herein.

[0050] [Figure 6A] This specification presents exemplary enumerations and pattern estimators for determining the probability that a hypothetical RNA sequence from a haplotype exhibits a unique mutation pattern and transcript set combination, according to several embodiments disclosed herein.

[0051] [Figure 6B] For example, we present a schematic diagram illustrating the usefulness of determining the incidence of haplotype transcripts when selecting tumor-associated peptides or proteins for the development of personalized anti-cancer therapies.

[0052] [Figure 7] The following are illustrative diagrams of statistical models for phasing tumor haplotypes using RNA and / or DNA sequence data, according to some embodiments disclosed herein.

[0053] [Figure 8A] Performance evaluation data from simulations using the techniques described herein, based on several embodiments disclosed herein, are shown.

[0054] [Figure 8B] Performance evaluation data from simulations using the techniques described herein, based on several embodiments disclosed herein, are shown.

[0055] [Figure 8C] This section presents performance evaluation data for estimated φh values ​​(using the posterior mean as the estimator) that demonstrate close agreement with ground truth φh values ​​in NA12878 genomic DNA.

[0056] [Figure 8D] Performance evaluation data for using the disclosed method to perform binary classification of haplotypes in NA12878 genomic DNA are presented.

[0057] [Figure 8E] Performance evaluation data for using the disclosed method to perform binary classification of haplotypes in NA12878 genomic DNA are presented.

[0058] [Figure 9] The following are flowcharts of methods for carrying out one or more methods for phasing mutations in a target tumor, according to some embodiments disclosed herein.

[0059] [Figure 10] This specification illustrates exemplary computing systems according to several embodiments disclosed herein. [Modes for carrying out the invention]

[0060] Description of Exemplary Embodiments I. Overview

[0061] Phasing analysis limited to normal tissues—reflecting a simpler issue than tumor phasing—involves separating the maternal and paternal copies of each chromosome into two haplotypes in order to obtain a clearer overall picture of the genetic variation of a given subject. A haplotype refers to a set of genomic variants along a single chromosome that tend to be inherited together. Phasing analysis is required because, as shown below, the parental labeling of a given sequence read (i.e., whether the read originates from the maternal or paternal copy of the gene) is not directly discernible in the sequencing data of the subject.

[0062] Figures 1A and 1C illustrate various challenges in exemplary phasing analysis. In Figure 1A, multiple sequence reads 102 (e.g., DNA sequence reads) may be obtained from the subject, some of which contain genomic variants. However, multiple sequence reads 102 represent an ambiguous overall picture of the subject's genetic variation. For example, read 102 may result from Scenario 1, which includes a maternal haplotype with a first genetic variant and a paternal haplotype with a second genetic variant. However, instead, read 102 may also result from Scenario 2, which includes a normal maternal haplotype and a paternal haplotype with both the first and second variants. In the example in Figure 1A, the ambiguity can be resolved because only Scenario 2 can produce sequence read 104, as the hypothetical sequence read 104 containing both the first and second genomic variants can be resolved. This is called readback phasing, which will be explained in more detail below.

[0063] Figure 1B illustrates the tumor mutation phasing problem (with normal contamination) in several embodiments. During bulk sequencing of the tumor in question, both the tumor cells in question and several normal cells are sequenced. Tumor cells contain somatic and germline variants, while normal cells may contain germline variants. Thus, as shown in the example in Figure 1B, bulk sequencing of a tumor may yield three or more haplotype-generating reads, e.g., a maternal normal haplotype, a paternal normal haplotype (containing germline variants), and a paternal mutant haplotype (containing both germline and somatic variants). As shown in Figure 1B, three haplotypes result in a complex mixture of sequence reads (e.g., DNA sequence reads), making it more difficult to identify the underlying haplotype based on the sequence reads.

[0064] Figure 1C illustrates the tumor mutation phasing problem with subclones in several embodiments. Figure 1C shows five possible haplotypes of the subject, including the maternal normal haplotype, paternal normal haplotype, paternal mutant haplotype, first maternal mutant haplotype, and second maternal mutant haplotype. The five haplotypes, here again, result in a complex mixture of sequence reads (e.g., DNA sequence reads), making it more difficult to identify the underlying haplotype based on the sequence reads.

[0065] Figure 1D illustrates the challenges in exemplary read-back phasing analysis of DNA or RNA sequence reads. As shown in Figure 1D, in the case of multiple mutations, most sequence reads will not cover all mutations. This may be because the sequence reads are too short, mutations occur within paired terminal read insertion regions, and / or the aligned sequence reads are simply located where no mutations occur. Furthermore, sequence reads can be affected by sequencing errors. In the example in Figure 1D, sequence read 112, showing 0 at the first base position, 1 at the second base position, and 0 at the third base position (1 indicates the presence of an alternative allele, and 0 indicates its absence), may represent the underlying haplotype A. Sequence read 114, showing 1 at the first base position, 1 at the second base position, and 0 at the third base position, may represent the underlying haplotype B. However, the majority of reads do not clearly indicate the presence of the underlying haplotype due to the problems described above.

[0066] Figure 1E shows an exemplary read-back phasing analysis for RNA sequence reads. RNA transcripts are single-stranded ribonucleic acid molecules synthesized from a DNA template during transcription. Some of these RNA transcripts can code for proteins, and before processing, they are known as precursor mRNA molecules. As part of the process of converting precursor mRNA to mature mRNA, regions known as "introns" are spliced ​​by spliceosome molecules, leaving behind regions known as "exons." Some of these exon regions code for the final protein product after translation. Importantly, precursor mRNA can be spliced ​​in multiple (but finite) ways, with some transcripts retaining one subset of exons and others retaining different subsets of exons. Also, a region that is an intron in one transcript may be an exon in another transcript. Each subset of exons known to likely remain after splicing is called a transcription isoform. On average, each protein-coding gene is associated with about four known transcription isoforms. Also, the transcription isoforms of a given gene typically share exons with each other. Finally, while “transcript” is typically a term used to describe individual RNA molecules—more precisely, the pattern of exons that can remain after splicing, which is called a “transcript isoform”—it should be noted that transcript isoforms will simply be referred to as transcripts in this document. Any reference to a transcript molecule will involve the explicit use of the word “molecule.”

[0067] In some embodiments, mature mRNA sequence read data are used to implement the techniques described herein to improve mutation phasing. Mature mRNA sequencing detects and quantifies the expression of transcript sequences in which introns have already been spliced. This is in contrast to whole RNA sequencing, which can detect and quantify the expression of immature transcripts. Mature mRNA data may be more relevant because it phases only mutations that have not been spliced ​​and are likely to eventually be translated into peptides, i.e., potential neoantigen targets. It should be understood that the techniques described herein can also be used to phase mutations found in whole RNA sequencing, for example, by using different immature transcript models.

[0068] Figure 1E shows two exemplary RNA transcripts. As shown in Figure 1E, each of RNA transcripts t1 and t2 contains two exons linked to an intron. One of the exons is shared between both transcripts.

[0069] For a given gene, multiple discontinuous transcript "regions" can be defined. Each region is also described as corresponding to a specific "transcript group," which is one of the subsets of transcripts expressed by the gene. A region is defined as corresponding to a transcript group if all the bases within that region are part of an exon expressed by all transcripts of that transcript group, and are not expressed by any transcripts outside of that transcript group, allowing all locations within a gene's exon to be classified as part of a single, unique transcript group. In general, for a given gene, a system (e.g., the enumeration and pattern probability estimator 260 shown in Figure 2B) can divide it into regions exclusive to transcript t1, another region exclusive to transcript t2, ... another region exclusive to transcript tj, and for every pair j and k, it can divide it into discontinuous regions shared only by tj and tk, and repeat for every triplet (j, k, l) for every j!=k, and so on, until the entire gene sequence (i.e., all base positions in all exons of the gene) is classified to belong to one of these discontinuous regions. In other words, transcript groups can be defined for all single genes (transcript groups containing a single transcript), pairs (transcript groups containing pairs of transcripts), triplets (transcript groups containing three transcripts), quads (transcript groups containing four transcripts), ... and so on. Notably, a single transcript can produce reads belonging to multiple transcript groups. Some exons within a transcript are shared among one set of transcripts, while other exons within a transcript are shared among different sets of transcripts. In the example shown in Figure 1E, for two transcripts, there is one region 402 exclusive to transcript t1, a second region 406 exclusive to transcript t2, and a third region 404 shared by t1 and t2. In another example, if there are three transcripts (t1, t2, t3), there are seven transcript groups corresponding to (t1), (t2), (t3), (t1, t2), (t1, t3), (t2, t3), and (t1, t2, and t3).

[0070] The potential transcript set can be enumerated by selecting all subsets (excluding empty subsets) of all possible transcripts that overlap with a given gene or gene region. Thus, if a given gene or gene region corresponds to three transcripts as described in the previous example (the number of transcripts originates, for example, from an existing gene transfer format GTF file), then N=(2 3 -1) There are sets of transcripts (i.e., in this case again, they correspond to (t1), (t2), (t3), (t1,t2), (t1,t3), (t2,t3), and (t1,t2,t3). Once a set of transcripts is enumerated, a given sequence read can be assigned to a specific set of transcripts by performing a step that includes, for example, the following: (i) for each base in the sequence read, identify one or more transcripts that have that base at the genomic coordinates shown in the GTF file above (if there are no transcripts with that base at those genomic coordinates, a transcript cannot be identified for the sequence read); (ii) the base is found in two or more transcripts of the GTF If a sequence read is part of another, it is assigned to the transcript group that includes each of those transcripts; if it does not include any other transcripts, and (iii) the entire sequence read is assigned to the transcript group that is smallest for any base in the sequence read. For example, assume the following: 80% of the sequence reads contain bases that can be assigned to transcript 1, transcript 2, or transcript 3; 5% of the sequence reads contain bases that may be part of transcript 1 or transcript 2; 15% of the sequence reads contain bases that can be assigned to transcript 1 only. In this case, since transcript 1 itself is the smallest transcript group, the sequence read is assigned to transcript group 1.

[0071] Readback phasing can be performed to assign each RNA sequence read to a single, unique group of transcripts; however, such transcript group assignment (or labeling) is, by construction, ambiguous with respect to the assignment of RNA sequence reads to a particular single transcript. In the example in Figure 1E, each read labeled g1 can only be obtained from transcript t1 because it aligns at least partially to exon 402, a region unique to transcript t1. Similarly, each read labeled g3 can only be obtained from transcript t2 because it aligns at least partially to exon 406, a region unique to transcript t2. However, the read labeled g2 aligns to region 404, which is common to both transcript t1 and transcript t2, thus introducing ambiguity that phasing analysis must consider.

[0072] Disclosed herein are exemplary devices, apparatus, systems, methods, and non-temporary storage media for phasing mutations in a tumor of interest. An exemplary system may access sequence reads derived from tumor cells obtained from the subject. Sequence reads may include tumor DNA sequence reads and / or tumor RNA sequence reads. According to the techniques described herein, a system may analyze sequence reads using a novel statistical model of tumor sudden phasing to estimate, for one or more haplotypes, one or more of the following: a set of probabilities of each haplotype existing (also called haplotype existence probabilities), a set of haplotype incidence rates, and a set of haplotype transcript incidence rates (the incidence rates of molecules from a particular haplotype that are simultaneously encoded by a particular transcript).

[0073] For example, for haplotype A, the system can estimate the posterior probability distributions of the events in which haplotype A is present, the incidence of haplotype A, and / or the incidence of one or more haplotype transcripts associated with haplotype A. These estimates of posterior probability distributions can, in principle, be used in their raw, sample-based form in subsequent analysis (among other benefits, potentially informing subsequent steps of the degree of uncertainty), or they can be summarized into point estimates (by mean, median, or some other statistical value) of the incidence of haplotype A expressed in each transcript for the entire set of RNA transcript molecules, where RNA sequence data is available. Accordingly, embodiments of the present disclosure can analyze DNA and / or RNA sequence reads derived from tumor cells obtained from a subject, assign which underlying haplotypes are represented in the sequence read data (by matching the sequence reads to unique mutation patterns identified in the sequence read data), and perform statistical analysis (e.g., using a statistical model) to generate a set of posterior probability distributions for i) the presence of each haplotype, and / or ii) the incidence of each haplotype, and / or iii) the incidence of each haplotype transcript combination. In some cases, one or more of these posterior probability distributions can be used directly in downstream analysis. In some cases, these posterior probability distributions can be used to calculate summary statistics (e.g., mean, median, or other statistics) as point estimates for use in downstream analysis.

[0074] Accordingly, embodiments of the present disclosure can model sequence data derived from tumor samples, which may include normal cells, polyclonal tumor cells, copy number variations, and other features that may cause the assumptions of the other phasing analysis tools described above to be rejected. Embodiments of the present disclosure can utilize tumor-normal-RNA-seq (TNR) triplet data from the same patient to simultaneously model DNA and RNA. Furthermore, embodiments of the present disclosure can jointly model haplotype incidence and haplotype transcript incidence.

[0075] Furthermore, embodiments of the present disclosure can estimate the number of haplotypes present in a bulk sample (e.g., the number of haplotypes per gene) without assuming that the number of haplotypes is known. Existing methods may operate under the assumption that a fixed number of haplotypes are present in a sample. However, in reality, a sample may contain an arbitrary number of haplotypes. Embodiments of the present disclosure do not assume, or operate under, a fixed number of haplotypes. In fact, embodiments of the present disclosure can be used to estimate the number of haplotypes present in a sample (e.g., by estimating the probability of each of several possible haplotypes being present). In some embodiments, the system may output a list of haplotypes present in the sample (e.g., a list of haplotypes associated with their probability of presence above a given threshold).

[0076] An exemplary system can quantify parameter uncertainties, incorporate prior information about possible haplotypes, and estimate transcript-specific haplotype incidences. In some cases, embodiments of the present disclosure can estimate posterior probability distributions of haplotype incidences and posterior probability distributions of haplotype transcript incidences, rather than determining point estimates of these parameters.

[0077] Embodiments of the present disclosure can be used to accurately model proteins and develop patient-specific cancer therapies. For example, embodiments of the present disclosure can be used in a pipeline to characterize and rank candidate neoantigens for use in vaccine development. As described below, the pipeline may use a combination of tumor DNA sequencing data, normal DNA sequencing data (e.g., from peripheral blood mononuclear cells (PBMCs) or tumor-adjacent tissue), and / or tumor RNA sequencing data, in which reads may be short or long, and the DNA sequencing may be WES, WGS, sWGS, tumor-targeted sequencing, or any other similar DNA sequencing approach, and the tumor RNA sequencing may be ribosome-depleted RNA sequencing, poly(A) mRNA sequencing, amplicon sequencing, hybridization capture sequencing, total RNA sequencing, or any other similar RNA sequencing approach.

[0078] Embodiments of this disclosure can be used to perform phasing analysis to accurately estimate the presence and incidence of various haplotypes, which can then be used to identify tumor mutation proteins (neoantigens). For example, a set of haplotypes identified by sequence read data (e.g., based on a comparison of haplotype probabilities with a given threshold) can be translated in silico to determine tumor-associated peptide or protein sequences (e.g., by parsing nucleotide sequence data into codons and then translating the codons into corresponding amino acid sequences). Haplotype incidence data output by the statistical model can be used to rank and / or select subsets of tumor-associated peptide or protein sequences for use in the development of patient-specific therapies (e.g., anti-cancer therapies). Haplotype transcript incidence data output by the statistical model can be used to further refine the ranking and / or selection of subsets of tumor-associated peptide or protein sequences for use in the development of patient-specific therapies (e.g., anti-cancer therapies) by determining, for example, that one tumor-associated peptide or protein transcript is far more abundant than another tumor-associated peptide or protein transcript.

[0079] A pipeline can be used to characterize and select patient-specific neoantigens (e.g., based on the predicted immunogenicity of the protein) that are targeted by the vaccine. In some embodiments, the pipeline can identify and / or select a predetermined number of neoantigens for each patient (e.g., 5, 10, 20, or more than 20 neoantigens can be selected for each patient) and identify personalized neoantigen-specific therapies for inducing an immune response in a given patient. In some embodiments, neoantigen-specific therapies can induce priming by RNA vaccination, DNA vaccination, injection of engineered and primed T cells and primed T cells, or any combination thereof.

[0080] Embodiments of this disclosure may be useful in any use case, including the modeling of tumor proteins. Depending on the indication, it is estimated that up to 10% of somatic missense mutations are nearly concentric mutations and are currently ignored without phasing analysis. Furthermore, previously published results suggest that the lack of proper phasing analysis can result in a 7% false-positive rate predicting p-MHC binding, and that the predicted binding affinity can be reduced 70-fold by the presence of one concentric and proximal germline mutant. In addition, priming for the wrong peptide (e.g., a non-existent peptide) may prime ineffective T cells in the context of antigen-focused immunotherapy. More broadly, embodiments of this disclosure may be useful whenever it is necessary to accurately model tumor proteins, as proteins resulting from different haplotypes may have different properties, different structures, and different immunogenicities.

[0081] In some embodiments, to estimate the above probability and incidence values, the exemplary system can determine the number of inputs to a statistical model based on sequence reads and provide those inputs to the statistical model. For example, the exemplary system can identify and enumerate a set of unique mutation patterns (i.e., all possible combinations of observed mutant alleles that can potentially be revealed by read alignment within a gene or gene region) for a set of mutations input to the system. A mutation pattern represents a particular combination of mutant alleles. One exemplary unique mutation pattern could be "1,0,1", which represents an allele value of "1" at the first base position, an allele value of "0" at the second base position, and an allele value of "1" at the third base position, where 1 indicates the presence of an alternate allele and 0 indicates the presence of a wild-type allele. Another exemplary unique mutation pattern could be "1,1,1", which represents an allele value of "1" at the first base position, an allele value of "1" at the second base position, and an allele value of "1" at the third base position. For each unique mutation pattern identified, the system can further count the number of times that unique mutation pattern was actually observed in sequence reads. For example, the system can count that the unique mutation pattern "1,1,1" appeared in two sequence reads, and therefore determine that the amount of the mutation pattern "1,1,1" is 2. In some embodiments, a mutation pattern may contain "?" at a given base position (i.e., the allele is not observed by a particular read at a given base position) because many sequence reads in a short read phasing may not cover all the mutations present in a given gene. Therefore, the system can enumerate one or more unique mutation patterns and, for each unique mutation pattern, determine the number of associated mutation patterns (e.g., by counting the number of sequence reads exhibiting a given unique mutation pattern).

[0082] As a second input provided to the statistical model, an exemplary system can determine, for each unique mutation pattern, the probability that a sequence read from a particular haplotype exhibits a particular mutation pattern for each haplotype in a set of candidate haplotypes. Such probabilities can be determined, for example, by performing a weighted sum of the number of sequence reads that would exhibit a particular mutation pattern by shifting the ends of a sequence read of a specified length along the haplotype sequence one base at a time, based on the position of the haplotype variant, the read's length, and its relative position to the candidate haplotype sequence (for example, by enumerating all possible sequence reads that can be generated relative to the haplotype sequence that match the mutation pattern, each weight multiplied by the probability of the required amount of sequencing error required for the candidate haplotype to generate that sequence read, and dividing this weighted sum by the total weighted sum of all possible sequence reads that can be generated by a particular haplotype sequence, regardless of the mutation pattern, by the insert of that read pair. This represents the probability of a given length. For example, the system can determine a first probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A will exhibit the unique mutation pattern "1,0,1". Furthermore, the system can determine a second probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A will exhibit the unique mutation pattern "1,1,1". Furthermore, the system can determine a third probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A will exhibit the unique mutation pattern "1,1,?", where "?" indicates the location of a mutation not covered by a given individual read, where the allele at that location is not determined. Thus, multiple probabilities can be determined for a given haplotype corresponding to multiple unique mutation patterns.

[0083] As a third input provided to the statistical model, the exemplary system can enumerate combinations of potentially unique mutation patterns and unique transcript sets found in RNA sequence reads. As a fourth input provided to the statistical model, the exemplary system can determine the probability that, for each combination of candidate haplotypes and transcripts of a given gene, a hypothetical RNA sequence read from that haplotype and transcript exhibits a given unique mutation pattern and unique transcript set. For example, for a combination of haplotype A and a first transcript, the system can determine the probability that a hypothetical RNA sequence read from that haplotype exhibits the unique mutation pattern "1,0,1" in a given combination of haplotype A and the first transcript set, and the probability is determined in a similar manner to that described above for DNA sequence reads, the main difference being that the count is performed conditionally on specific haplotypes and specific transcripts. As another example, the system can determine the probability that a hypothetical RNA sequence read from haplotype A exhibits a unique mutation pattern "1,0,1" in a given combination of haplotype A and some other group of second transcripts.

[0084] An exemplary system may input the quantity and probability of mutation patterns into a statistical model to estimate one or more of the following: i) the probability that each of a set of haplotypes exists (also called the probability of haplotype presence), ii) a set of haplotype incidence rates, and / or iii) a set of haplotype transcript incidence rates. In some embodiments, the statistical model includes a hierarchical Bayesian model (i.e., a statistical model described in multiple values ​​(hierarchical form) that uses Bayesian estimation to estimate parameters of a posterior probability distribution and updates the probabilities as additional data is provided). As shown in Figure 7 and described below, a hierarchical Bayesian model may include several components, e.g., a haplotype generation model used to determine which haplotypes are present in sequence read data, a DNA Dirichlet-Multinomial model used to determine haplotype incidence rates from DNA sequence read data, and / or an RNA Dirichlet-Multinomial model used to determine haplotype transcript incidence rates from RNA sequence read data. Those skilled in the art will understand that the techniques described herein can be performed using other statistical models.

[0085] Model Description Further explanation of the exemplary model, including alternative statistical models that can be used, is provided below.

[0086] Fading window model Fading window specific model structure (as a preprocessing step for the fading window specific model, which simultaneously considers sequence reads from the entire genome, as described later) TIFF2026520346000002.tif5170 and (In contrast to the model for TIFF2026520346000003.tif5170) enables the integration of haplotype sampling models, DNA sampling models, and RNA sampling models.

[0087] Haplotype Model Haplotypes can be modeled as follows:

number

number

number

[0088] II. Platform Overview

[0089] Figure 2A shows an exemplary process 200 for phasing mutations in a patient's tumor according to embodiments of the present disclosure. In certain embodiments, the workflow system 200 may include a combination of laboratory workflows and computing workflows. Referring to Figure 2A, as part of the laboratory workflow, one or more biopsies of one or more samples may be performed on the patient 210. One or more samples may include healthy samples from healthy tissue, tumor samples from tumor tissue, or a combination thereof. Furthermore, the tumor samples may include normal (i.e., healthy) tissue or cells 206, cancerous tissue or cells 208, or a combination thereof.

[0090] The laboratory workflow may further include determining DNA sequences 212 and RNA sequences 214 from one or more samples. For example, a tumor sample can be obtained from a patient 210, and DNA and / or RNA can be isolated from the tumor sample using any method known in the art. The DNA and / or RNA sequences can also be determined by any sequencing method known in the art. In some embodiments, next-generation sequencing methods, including large-scale parallel sequencing, are used to obtain short sequences, typically 50 to 400 base pairs in length. In some embodiments, the DNA template to be sequenced can be obtained by clonal emulsion PCR or clonal bridge amplification. In certain embodiments, the laboratory workflow may include a hybrid capture subprocess in which one or more sequence reads are generated based on the sequencing of DNA and / or RNA extracted from one or more samples. "Hybridization capture" refers to a targeted next-generation sequencing method that enriches a target genome sequence through hybridization of a tagged (e.g., biotinylated) bait oligonucleotide to a region of interest on a DNA fragment, and captures the bait oligonucleotide via tagging (e.g., using streptavidin-conjugated magnetic beads) before sequencing. In some embodiments, third-generation sequencing methods can be used to obtain long-read sequences typically exceeding 10 to 1000 kilobases in length. In certain embodiments, third-generation sequencing methods developed by Oxford Nanopore Technology can be used. In other embodiments, third-generation sequencing methods developed by Pacific Biosciences, including single-molecule real-time sequencing, can be used. In certain embodiments, one or more sequence reads generated by sequencing DNA and / or RNA extracted from one or more samples may include tumor DNA sequence reads, tumor RNA sequence reads, normal DNA sequence reads, or a combination thereof.

[0091] In one embodiment, a sequencing subprocess can generate raw base call data as a binary base call (BCL) file, which can then be converted and mapped to one or more sorted binary alignment map (BAM) files. For example, each sorted BAM file may contain data for one or more sequence reads (e.g., tumor DNA sequence reads, tumor RNA sequence reads, and normal DNA sequence reads) generated by sequencing DNA and / or RNA extracted from one or more samples. In a particular embodiment, the computing platform 202 can receive sorted BAM files containing base call data for sequence reads (e.g., tumor DNA sequence reads, tumor RNA sequence reads, and normal DNA sequence reads), and further use the sorted BAM files of sequence read data to perform one or more genomic variant calls (e.g., germline variant calls, somatic variant calls) to identify germline and / or somatic variants.

[0092] In certain embodiments, as further shown in Figure 2A, the computing platform 202 may include one or more computing devices (e.g., one or more servers and / or client devices) and one or more databases (e.g., data stores, relational databases). For example, in some embodiments, the computing platform 202 may include a cloud-based computing architecture suitable for performing techniques for phasing tumor mutations in patient 210 according to embodiments of the present disclosure. For example, in one embodiment, the computing platform 202 may include a Platform as a Service (PaaS) architecture, a Software as a Service (SaaS) architecture, an Infrastructure as a Service (IaaS) architecture, a Computation as a Service (CaaS) architecture, a Data as a Service (DaaS) architecture, a Database as a Service (DbaaS) architecture, or other similar cloud-based computing architectures (e.g., "X" as a Service (XaaS)).

[0093] The computing platform 202 can perform techniques for phasing mutations in a target tumor, as described herein. In certain embodiments, the computing platform 202 can utilize a statistical model (e.g., a hierarchical Bayesian model) to estimate one or more sets of probabilities (also referred to as sets of haplotype probabilities) that each of several haplotypes is present in the sample, based on sequence reads and identified germline and / or somatic mutations, as further illustrated below with respect to Figure 2B. These probabilities are demonstrated by sequence reads from the sample (e.g., tumor DNA sequence reads, tumor RNA sequence reads, and / or normal DNA sequence reads), a set of haplotype occurrences (from DNA and / or RNA sequence read data), and a set of haplotype transcript occurrences (from RNA sequence read data). In some embodiments, tumor DNA sequence reads and normal DNA sequence reads may be aligned to a reference genome (e.g., using BWA and STAR aligners, respectively) and compared to what are called somatic mutations (e.g., using MuTect or VarSim callers).

[0094] In certain embodiments, as further shown in Figure 2A, the exemplary system can determine one or more personalized cancer immunotherapy treatments 224 based on mutation phasing in sequence read data. For example, in some embodiments, the computing platform 202 may generate predictions of tumor neoantigens, which may include a set of mutant peptide sequences derived from expressed somatic mutations in sequence read data associated with a patient 210. Specifically, in some embodiments, phasing mutations can be translated in silico into a set of tumor neoantigens (e.g., mutant peptide sequences). In some embodiments, the system can determine which haplotypes are present based on the probability of haplotype presence (e.g., by determining whether the probability is non-zero or exceeds a predetermined threshold). These present haplotypes can be translated into peptide sequences without using incidence rates. In other words, the system can restrict downstream analysis of known present haplotypes based on sequence read data to avoid targeting absent haplotypes because they are not effective as treatments. The estimated incidence rates (e.g., haplotype incidence and / or haplotype transcript incidence) are used to select one or more corresponding peptide sequences (e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or more than 10 corresponding peptide sequences) to target for use in a vaccine or pharmaceutical composition. For example, if the incidence estimate for a given haplotype is greater than a threshold, the corresponding peptide sequence can be selected for further downstream processing. On the other hand, if the incidence estimate for a given haplotype is lower than a threshold (e.g., showing a very low incidence), the system can exclude the haplotype (and its corresponding peptide sequence) from downstream processing to avoid targeting neoantigens associated with low-incidence haplotypes in the vaccine.Alternatively, peptide sequences may be ranked at least partially according to the incidence of associated haplotype transcripts, and several subsets of those peptide sequences (e.g., peptide sequences 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or more than 10) may be selected for further downstream processing according to their rank. In some embodiments, one or more peptides corresponding to haplotypes identified in a sample can be selected using one or more predetermined selection criteria. Each of the one or more predetermined selection criteria may be applied to the estimated haplotype incidence, estimated haplotype transcript incidence, predicted likelihood of MHC binding, presentation, or immunogenicity, etc.

[0095] For example, in some embodiments, a selected set of candidate tumor neoantigens (e.g., mutant peptide sequences) can then be input into one or more machine learning models (e.g., one or more MHC binding and / or MHC presentation prediction models) to generate predictions of the likelihood of MHC presentation for each of the candidate tumor neoantigens, or predictions of immunogenicity for each of the candidate tumor neoantigens. In certain embodiments, the predictions of the likelihood of MHC presentation for each of the candidate tumor neoantigens, or the predictions of immunogenicity for each of the candidate tumor neoantigens, can then be used to select and / or prioritize subsets of candidate neoantigens and / or determine one or more personalized cancer immunotherapy treatments 224. For example, in one embodiment, one or more personalized cancer immunotherapy treatments 224 may include recombinant neoantigen-specific T cells (e.g., primed T cells) or neoantigen vaccines (e.g., RNA vaccines or DNA vaccines).

[0096] In some embodiments, the prediction of presentation and / or immunogenicity can be performed using NetMHC technology (see, e.g., Nielsen et al. (2020), “Immunoinformatics: Predicting Peptide-MHC Binding”, Annu. Rev. Biomed. Data Sci. 3: 191-215), which can predict peptide binding to several different HLA alleles using an artificial neural network (ANN) and a weight matrix. Additionally or alternatively, predictions can be made based on the processes disclosed in, for example, U.S. Patent Application No. 17 / 378,651, entitled “ATTENTION-BASED NEURAL NETWORK TO PREDICT PEPTIDE BINDING, PRESENTATION, AND IMMUNOGENICITY,” and U.S. Patent Application No. 63 / 430,297, entitled “PREDICTION OF PEPTIDE PRESENTATION BY MAJOR HISTOCOMPATIBILITY COMPLEX MOLECULES,” the respective contents of which are incorporated herein by reference. For example, the system may have access to a set of peptide sequences characterizing a set of peptides, each peptide sequence in the set of peptide sequences being identified by processing disease samples from a subject, and may have access to immunoprotein complex (IPC) sequences identified for the subject's immunoprotein complex (IPC). The system may then process a set of peptide representations representing the set of peptide sequences using a first attention block of the initial attention subsystem of an attention-based machine learning model, and an immunoprotein complex (IPC) representation representing the IPC sequences using a second attention block of the initial attention subsystem, to generate an output that includes at least one of interaction predictions, interaction affinity predictions, or immunogenicity predictions for the corresponding peptide-IPC combinations.As another example, the system predicts amino acid-immunoprotein complex (IPC) interactions by accessing a set of amino acid sequences, each of which has been identified from at least one protein, and has access to identified immunoprotein complex (IPC) sequences for the IPC of interest. The system can then use one or more first processing blocks of the machine learning model's processing subsystem to process the set of amino acid sequence representations to generate a set of transformed amino acid sequence representations based on a set of element-focused scores representing the bonded cores of the set of amino acid sequence representations, each of which is generated based on one of the aforementioned amino acid sequences with a start-of-sequence (BOS) token appended. The system can then use a second processing block of the processing subsystem to process the IPC sequence representations to generate transformed IPC sequence representations, each of which is generated based on identified IPC sequences with a BOS token appended, and the set of amino acid sequence representations and the IPC sequence representations are processed in parallel. The system can generate a composite representation by combining each of the transformed BOS token representations of the set of transformed amino acid sequence representations with the transformed BOS token representation of the transformed IPC sequence representation. The system can determine one or more predicted amino acid-IPC interactions based on the composite representation.

[0097] Figure 2B shows an exemplary computing platform 202 for performing mutation phasing in a patient's tumor according to embodiments of the present disclosure. As shown in Figure 2B, the computing platform 202 can access data relating to a set of sequence reads 254 (e.g., in one or more sorted BAM files) generated by sequencing DNA and / or RNA extracted from one or more samples of patient 210. In some embodiments, the set of sequence reads 254 may include tumor DNA sequence reads, tumor RNA sequence reads, normal DNA sequence reads, or a combination thereof. In certain embodiments, the set of sequence reads 254 may be generated using one or more short-read sequencing techniques and therefore relatively short in length (e.g., compared to long-read sequence reads).

[0098] In certain embodiments, as further shown in Figure 2B, the computing platform 202 can access or determine multiple genomic variant calls 256 (e.g., germline variant calls, somatic variant calls) associated with a set of sequence reads 254 (e.g., tumor DNA sequence reads, tumor RNA sequence reads, and normal DNA sequence reads) to identify multiple germline and somatic variants 256. To determine the genomic variants 256, the variant calling process can be performed by the computing system 202 or an upstream system in the pipeline. Non-limiting examples of suitable variant calling algorithms include HaplotypeCaller and Mutect2 (both from the Genome Analysis Toolkit, Broad Institute, Cambridge, Massachusetts) for calling germline and somatic variants, respectively, from DNA sequence read data. Non-limiting examples of variant calling algorithms suitable for use with RNA sequence read data include Mpileup / Varscan and Haplotype Caller from SAMtools (https: / / www.htslib.org / ) (e.g., after processing BAM files with SplitNCigarReads (Genome Analysis Toolkit, Broad Institute, Cambridge, Massachusetts)).

[0099] The mutation phasing described herein may be performed independently on subsets of mutations at one time. The disclosed method determines which mutations need to be phased together for accurate neoantigen determination by defining a phasing window (as described below) based on the span lengths assigned to different types of mutations. This determination of the phasing window (and span length) then depends on the length of the specified target neoantigen (e.g., approximately 21–27+ amino acid residues (allowing binding peptide + lateral modeling) or longer (i.e., approximately 63–81 bases or longer nucleic acid sequences) for peptides presented by MHC class I molecules, or approximately 15–34+ amino acid residues (i.e., approximately 45–102 bases or longer nucleic acid sequences) when targeting MHC class II neoantigens), and also on the type of upstream mutation identified. With respect to mutation type, accurate identification of neoantigens induced by frameshift indels or stop-loss mutations may require phasing mutations located several hundred bases apart from each other. For example, mutations found in non-overlapping genes do not need to be phased together to infer a possible neoantigen sequence, and thus different genes can be separated into different phasing windows. Since some subprocesses (e.g., Bayesian estimation) scale non-linearly as the number of mutations increases, splitting the phasing problem into a series of smaller problems further increases the computational feasibility of the method (while maintaining a phasing window large enough to predict neoantigens). Furthermore, even within the same gene, more distantly separated mutations may be difficult to phase in short-read sequencing data and are less likely to affect the sequence of a given relatively short (e.g., 21-27 amino acid lengths) neoantigen.

[0100] To minimize the number of variants (or mutations) phased at once, the system can first decompose the genome into phasing windows. For the first approximation, the phasing windows correspond to gene sequences. In some cases, the system may include multiple genes with overlapping genomic coordinates in the same phasing window. Subsequently, the system can further decompose the phasing windows into phasing subwindows based on the relative positions of variants identified in DNA and / or RNA sequence read data. The phasing windows are defined by iteratively adding the length of a specified span for each variant to a given “phasing window” definition (as described later) until the phasing windows stop overlapping with adjacent spans. Variants that are far apart from each other in a given gene (e.g., those associated with distance-separated genomic coordinates exceeding a specified distance threshold (e.g., a distance threshold of values ​​in the range of approximately 63 to 81 bases) and therefore exist in different phasing windows often do not need to be phased together to infer the sequence of a short neoantigen (e.g., a peptide sequence with a length of 21 to 27 amino acids). In some cases, for example, the system can determine the variant coordinates in transcript space along with the potential consequences of the variants. Each somatic variant coordinate (e.g., centered on the somatic variant coordinate) is assigned a span of user-specified length, and somatic variants with overlapping spans may be assigned to the same phasing window or sub-window. The system can assign variants (somatic or germline) without overlapping spans to different phasing windows. For the purpose of neoantigen phasing, the system may assign germline variants a span of 1 nucleotide (i.e., to prevent unnecessarily expanding the phasing window), except in the following special cases:Frameshift mutants, or mutants that cause stop codon loss or gain, whether germ cell or somatic cell origin, may be assigned an unbalanced span, with the downstream end of the span set at the 3' end of the transcript and the upstream end determined by the user-specified length of the span. This is due to the fact that the translation of downstream mutants of these special-case mutants may depend on whether they are homeophase with the frameshift / stop loss / gain mutants.

[0101] As further illustrated by Figure 2B, the computing platform 202 may include a sequence read counter (or read pattern counter) 258. The sequence read counter 258 can identify unique mutation patterns observed in sequence reads 254 and quantify or count each unique mutation pattern observed in sequence reads 254. Mutation patterns indicate specific combinations of mutant alleles (which may include “unobserved” alleles). One exemplary unique mutation pattern might be “1,0,1”, which indicates an allele value of “1” at the first base position, an allele value of “0” at the second base position, and an allele value of “1” at the third base position. Another exemplary unique mutation pattern might be “1,1,1”, which indicates an allele value of “1” at the first base position, an allele value of “1” at the second base position, and an allele value of “1” at the third base position. The letter “?” is used to indicate when no mutant state is observed in a given read. For each unique mutation pattern identified, the system can further count the number of times that unique mutation pattern is observed in sequence reads. For example, the system can count that the unique mutation pattern "1,1,1" appeared in two sequence reads, and therefore determine that the amount of the mutation pattern "1,1,1" is 2. Thus, the system can enumerate one or more unique mutation patterns (by generating all possible combinations of mutations identified in the sequence read data) and, for each unique mutation pattern, determine the amount of the associated mutation pattern (by counting the number of sequence reads (DNA and / or RNA sequence reads) that exhibit a given unique mutation pattern. Similarly, the system can enumerate a combination of a unique mutation pattern and a set of transcripts observed in RNA sequence reads (as described above) and determine the amount of the associated mutation pattern-transcript set (by counting the number of RNA sequence reads assigned to a specified set of transcripts exhibiting a given unique mutation pattern). Details of the sequence read counter are provided below with reference to Figures 3-4.

[0102] As further illustrated by Figure 2B, the computing platform 202 may include an enumeration and pattern probability estimator 260. For each unique mutation pattern, the enumeration and pattern probability estimator 260 determines the probability that a hypothetical DNA sequence and / or RNA sequence read from each of the set of haplotypes exhibits the unique mutation pattern. For example, the system may determine a first probability indicating the likelihood that a hypothetical DNA sequence read and / or RNA sequence read from haplotype A exhibits the unique mutation pattern "1,0,1". Furthermore, the system may determine a second probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A exhibits the unique mutation pattern "1,1,1". Thus, multiple probabilities can be determined for a given haplotype corresponding to multiple unique mutation patterns (a given haplotype is obtained from multiple candidate haplotypes generated by taking all possible combinations of mutations in a gene or gene region and their associated alleles). Similarly, the system can determine the probability that a virtual RNA sequence from a particular haplotype and transcript pair exhibits a unique mutation pattern and transcript group labeling for each set of haplotype combinations and sets of transcript groups associated with those haplotypes (transcript group labeling indicates the transcript group to which the RNA sequence read is assigned). Details of the enumeration and pattern probability estimators are provided below with reference to Figures 5A–6.

[0103] As further illustrated by Figure 2B, the computing platform 202 may include statistical models 262, such as a hierarchical Bayesian model. Based on the quantity of mutation patterns / mutation pattern-transcripts (obtained by the sequence readout counter 258) and the probability of conditional mutation patterns of haplotypes / haplotype transcripts (obtained by the enumeration and pattern probability estimator 260), the statistical model 262 can be used to estimate an output 264 that includes one or more of the following: a set of probabilities of each haplotype existing (also called haplotype existence probabilities), a set of haplotype existence probabilities, and / or a set of haplotype transcript occurrence rates. Details of an exemplary model are provided below with reference to Figure 7.

[0104] As further shown in Figure 2, the system 200 can identify a set of peptide sequences using translations 266 of one or more haplotype transcripts based on a set of haplotype probabilities. In some embodiments, the system identifies a set of mutant peptide sequences based on a set of peptide sequences. In some embodiments, the system can determine which haplotypes are present based on haplotype probabilities (e.g., by determining whether the probability is non-zero or exceeds a given threshold). These present haplotypes can be translated into peptide sequences. In other words, the system can restrict downstream analysis for known present haplotypes based on sequence read data to avoid targeting absent haplotypes because they are not effective as treatments. Incidence is used to select a subset of peptide sequences to target in the vaccine. For example, if the incidence estimate for a given haplotype is greater than a threshold, the corresponding peptide sequence can be selected for further downstream processing. On the other hand, if the incidence estimate for a given haplotype is lower than a threshold (e.g., showing a very low incidence), the system can exclude the haplotype from downstream processes to avoid targeting low-incidence haplotypes in the vaccine. Alternatively, peptide sequences can be ranked at least partially according to their incidence, and several subsets of those peptide sequences can be selected for further downstream processing according to their rank.

[0105] In certain embodiments, one or more machine learning models (e.g., one or more MHC binding and / or MHC presentation prediction models) may generate predictions of the likelihood of presentation in MHC for one or more sets of mutant peptide sequences, or predictions of immunogenicity for one or more sets of mutant peptide sequences. The predictions of the likelihood of presentation in MHC for one or more sets of mutant peptide sequences, or the predictions of immunogenicity for one or more sets of mutant peptide sequences, can then be used to identify one or more candidate neoantigens and to determine one or more personalized cancer immunotherapy treatments, such as one or more genetically modified neoantigen-specific T cells (e.g., primed T cells) or neoantigen vaccines (e.g., RNA vaccines or DNA vaccines).

[0106] III. Array Read Counter

[0107] III. Sequence Read Counter - DNA Read

[0108] Figure 3 illustrates exemplary processing by an exemplary sequence read counter 258 for enumerating and quantifying unique variant patterns observed in sequence read data (e.g., DNA sequence reads) according to embodiments of the present disclosure. In these examples, the weighted counting process described below is omitted for simplification. The counter 258 accesses data for multiple aligned (DNA and / or RNA) sequence reads 302, which may be part of the sequence read data 254 and variant calls 256 in Figure 2A.

[0109] In the illustrated example, the sequencing data includes data from a pair of sequence reads. Sequencing may include single-ended (SE) reads or paired-ended (PE) reads. Paired-end sequencing involves sequencing both ends of a DNA fragment. The forward and reverse reads can then be aligned. As described herein, sequence reads may include a pair of two reads or a read pair, including a forward read and a reverse read, as shown in Figure 3.

[0110] Referring to Figure 3, the sequence read counter analyzes sequence read 302 to identify a unique mutation pattern 308. For example, multiple aligned (DNA and / or RNA) sequence reads 302 may exhibit multiple mutation patterns 306A–306V. For example, as shown in Figure 3, ● The aligned sequence read pair 306A and 306B exhibit the mutation pattern "0,1,0". ● The aligned sequence read pair 306C and 306D exhibit the mutation pattern "?,1,0". ● The aligned sequence read pair 306E and 306F exhibit the mutation pattern "?,?,0". ● The aligned sequence read pair 306G and 306H exhibit the mutation pattern "0,?,0". ● The aligned sequence read pair 306I and 306J exhibit the mutation pattern "0,?,0". ● The aligned sequence read pair 306K and 306L exhibit the mutation pattern "?,?,?". ● The aligned sequence read pair 306M and 306N exhibit the mutation pattern "?,1,1". ● The aligned sequence read pair 306O and 306P exhibit the mutation pattern "?,?,?". ● The aligned sequence read pair 306Q and 306R exhibit the mutation pattern "1,1,0". ● The aligned sequence read pairs 306S and 306T exhibit the mutation pattern "0,?,1". ● The aligned sequence read pair 306U and 306V exhibit the mutation pattern "0,?,1".

[0111] In the example shown in Figure 3, each time the sequence read counter identifies a unique mutation pattern (e.g., the pattern in column 308) in the aligned (DNA or RNA) sequence read 302, the sequence read counter may update table 304 (or database, or other preferred memory structure) to include the identified mutation pattern and the number of times the mutation pattern for the remaining reads for the relevant sample of interest is counted. For example, ● In the aligned read pair 306A-B, the mutation pattern "0,1,0" was observed once and therefore counted as 1. ● In the aligned read pair 306C-306D, the mutation pattern "?,1,0" was observed once and therefore counted as 1. ● In the aligned read pair 306E-F, the mutation pattern "?,?,0" was observed once and is therefore counted as 1. ●The mutation pattern "0,?,0" was observed twice in the aligned read pair 306G-H, and then in the aligned read pair 306I-J, and is therefore counted as 2. ● In the aligned read pair 306K-L, the mutation pattern "?,?,?" was observed once and therefore counted as 1. ● In the aligned read pair, the variation pattern "1,1,?" was observed once in the 306Q-R, and therefore counted as 1. ●The mutation pattern "0,?,1" was observed in the aligned read pair 306S-T, and then in the aligned read pair 306U-V, and is therefore counted as 2. For tumor DNA sequence reads, the number of sequence reads exhibiting each unique mutation pattern is counted. For tumor RNA sequence reads, the number of sequence reads exhibiting each combination of unique mutation pattern and transcript group is counted. The DNA and RNA counting operations are performed separately when both DNA and RNA sequence read data are available. In some embodiments, the number of normal tissue sample DNA sequence reads exhibiting each unique mutation pattern is also counted, and normal haplotype phasing is estimated using a DNA-only model. This data provides information on the normal proteome and can then be used to ensure that candidate tumor-associated neoantigens are truly tumor-specific targets. Furthermore, haplotypes identified in the normal data can be input into the tumor phasing model as a required part of the tumor phasing solution (to account for contamination).

[0112] III. Sequence Read Counter-RNA Read

[0113] According to some embodiments, a sequence read counter can also be used to count how many sequence reads are present in a particular group of RNA transcripts. Figure 4 shows an exemplary process of an exemplary sequence read counter 258 for enumerating and quantifying unique mutation patterns observed in RNA sequence reads according to embodiments of the present disclosure.

[0114] Referring to Figure 4, the sequence read counter analyzes sequence reads to identify unique mutation patterns and transcript combinations. In the illustrated example, multiple reads produce the following unique combinations of mutation patterns and transcripts. ●0,?,?.?;g1 ●0,0,?.?;g1 ●1,?,?,?;g1 ●?,1,?,?;g1 ●?,?,0,?;g1 ●?,?,1,?;g1 ●?,?,?,?;g2 ●?,?,?,1;g3 ●?,?,?,0;g3 ●?,?,?,?;g3

[0115] In the example shown in Figure 4, each time the sequence read counter identifies a unique combination of mutation pattern and transcript group, the sequence read counter updates table 400 (or database, or other preferred memory structure) to include the identified combination, and can track the occurrences of that identified combination. For example, ●The combination of the mutation pattern "0,?,?,?" and g1 was observed once and counted as 1. ●The combination of the mutation pattern "0,0,?,?" and g1 was observed once and counted as 1. ●The combination of mutation pattern "1,?,?,?" and g1 was observed once and counted as 1. ●The combination of the mutation pattern "?,1,?,?" and g1 was observed once and counted as 1. ●The combination of the mutation pattern "?,?,0,?" and g1 was observed once and counted as 1. ●The combination of the mutation pattern "?,?,1,?" and g1 was observed once and counted as 1. ●The mutation pattern "?,?,?,?" was observed twice in combination with g2, and was counted as 2. ●The combination of the mutation pattern "?,?,?,1" and g3 was observed once and counted as 1. ●The combination of the mutation pattern "?,?,?,0" and g3 was observed once and counted as 1. ●The mutation pattern "?,?,?,?" combined with g3 was observed twice and counted as 2.

[0116] IV. Enumeration and Pattern Probability Estimators

[0117] IV.A.DNA lead

[0118] The enumeration and pattern estimator, for each haplotype h, enumerates all possible sequence reads that can be generated for that haplotype (e.g., by shifting sequence reads of a given length one base at a time along the haplotype sequence) and performs a weighted sum of the number of sequence reads that would exhibit a given unique mutation pattern m (weights take into account sequencing error and insert length). Then, using the enumerated and weighted counted set of sequence reads, it calculates the probability W corresponding to the unnormalized molecule of the probability that a hypothetical DNA sequence read from that haplotype exhibits a unique mutation pattern. mh The system determines the probability of a unique mutation pattern (this quantity can be normalized by dividing the weighted number of sequence reads expected to exhibit a unique mutation pattern by the weighted total number of sequence reads that can be generated for a haplotype regardless of the mutation pattern; the normalization constant is unnecessary as it cancels out during the model calculation). Returning to Figure 2B, the enumeration and pattern probability estimator 260 determines the denormalized probability that a hypothetical DNA sequence read from each of the haplotype sets exhibits a unique mutation pattern for each unique mutation pattern. For example, the system can determine a first denormalized probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A exhibits the unique mutation pattern "1,0,1". Furthermore, the system can determine a second denormalized probability indicating the likelihood that a hypothetical DNA sequence read from haplotype A exhibits the unique mutation pattern "1,1,1". Thus, multiple denormalized probabilities can be determined for a given haplotype corresponding to multiple unique mutation patterns.

[0119] Figures 5A and 5B illustrate exemplary processing by an enumeration and pattern probability estimator 260 for determining the probability that a hypothetical DNA sequence from a given haplotype exhibits a given unique mutation pattern, according to embodiments of the present disclosure. The estimator accesses one or more theoretical exemplary haplotypes 502 ("H1") in Figure 5A and 504 ("H2") in Figure 5B (one or more theoretical exemplary haplotypes are a set of mutations identified in multiple sequence reads and a specific combination of their associated alleles). Exemplary haplotype 502(H1) may contain the allele "1" at the first base position a, "1" at the second base position b, "0" at the third base position c, and "1" at the fourth base position c, while exemplary haplotype 504(H2) may contain the allele "1" at the first base position a, "1" at the second base position b, "1" at the third base position c, and "1" at the fourth base position d. The estimator determines the probability that a hypothetical DNA sequence from a given haplotype exhibits a given unique mutation pattern, as described below.

[0120] As shown, estimator 260 can determine probability 510A that a hypothetical DNA sequence read from haplotype 502(H1) may exhibit a unique mutation pattern "1,1,0,1" in the sequence read. Probability 510A can be a relatively low probability indicating a low-probability event, since the sequence read pair must have an exact matching order and contain a read pair with sufficient positioning and / or length for the "1", "1", "0", "1" allele pattern present in exemplary haplotype 502(H1), given that a sequence read pair containing allele "1" at base position a, "1" at base position b, "0" at base position c, and "1" at base position d must be found. An example in Figure 5A illustrates this as a low-probability event by showing the only (in this example) possible order, position, and / or length of a hypothetical sequence read from capturing such an allele pattern.

[0121] As another example, estimator 260 can determine probability 510B, indicating that a hypothetical DNA sequence read pair from haplotype 502(H1) may exhibit a unique mutation pattern "1,1,0,?". Probability 510B may be higher than probability 510A, indicating a more likely event because haplotype 502 can yield read pairs exhibiting "1,1,0,?" for more positions. The "?" no longer covers the last mutation position at this point, giving the right-hand read of the pair greater freedom in its placement. An example in Figure 5A illustrates this as a relatively high probability event by showing that there are six hypothetical sequence read pairs with such order, position, and / or length to capture such an allele pattern.

[0122] As another example, the estimator can determine probability 510C, indicating that a hypothetical DNA sequence read pair from haplotype 502(H1) may exhibit a unique mutation pattern "1,1,1,1" in the sequence reads. Probability 510C may be lower than probability 510A, representing an extremely unlikely event, as it is impossible that a sequence read pair containing the allele "1" at base position a, "1" at base position b, "1" at base position c, and "1" at base position d could possibly match, in terms of both order and position, the "1", "1", "0", "1" allele pattern present in exemplary haplotype 502(H1) without any form of sequencing error in a possible but rare third allele. The example in Figure 5A illustrates this as a low-probability event by showing the only (in this example) possible order, position, and / or length of a hypothetical sequence read from capturing such an allele pattern.

[0123] Similarly, as further shown in Figure 5B, various probabilities can be determined by the estimator for the exemplary haplotype 504(H2). For example, the estimator can determine a probability 512A indicating that a hypothetical DNA sequence read pair from haplotype 504(H2) may exhibit a unique mutation pattern "1,1,1,1" in the sequence reads. When a sequence read pair containing the allele "1" at base position a, "1" at base position b, "1" at base position c, and "1" at base position d must be found in an order and position that precisely matches the "1", "1", "1", "1" allele pattern present in exemplary haplotype 504(H2), probability 512A may be a relatively low probability suggesting an unlikely event. The example in Figure 5B illustrates this as a low-probability event by showing the only (in this example) possible order, position, and / or length of the hypothetical sequence read for capturing such an allele pattern.

[0124] As another example, the estimator can determine probability 512B, indicating that a hypothetical DNA sequence read pair from haplotype 504(H2) is likely to exhibit a unique mutation pattern "1,1,1,?" in the sequence reads. Probability 512B may be higher than probability 512A, suggesting a more likely event, when a sequence read pair containing the allele "1" at base position a, "1" at base position b, "1" at base position c, and "?" at base position d is more likely to be found in an order and position that precisely matches at least three alleles of the "1", "1", "1", "1" allele pattern present in exemplary haplotype 504 (e.g., H2). The example in Figure 5B illustrates this as a relatively high probability event by showing that there are six hypothetical sequence reads with such order and position and / or length to capture such an allele pattern.

[0125] As another example, the estimator can determine probability 512C, indicating that a hypothetical DNA sequence read from haplotype 504(H2) may exhibit a unique mutation pattern "1,1,0,1" in the sequence read. Probability 512C may be lower than probability 512A, representing an extremely unlikely event, as it is impossible that a sequence read pair containing the allele "1" at base position a, "1" at base position b, "0" at base position c, and "1" at base position d could possibly match, in both order and position, the "1", "1", "1", "1" allele pattern present in exemplary haplotype 504(H2) without any form of sequencing error in a possible but rare third allele. The example in Figure 5B illustrates this as a low-probability event by showing the only (in this example) possible order, position, and / or length of a hypothetical sequence read for capturing such an allele pattern.

[0126] IV.B. RNA reads

[0127] In certain embodiments, the enumeration and pattern probability estimator 210 is further used to determine the denormalized probability W for each unique mutation pattern m that, when arising from the molecule and transcript t of haplotype h, the hypothetical RNA sequence from each combination of the set of haplotype-transcript combinations exhibits the unique mutation pattern m and transcript group g. gmht It is possible to determine the probability W. gmht This can be determined, for example, by enumerating all possible RNA sequence reads that can be generated for haplotype h and transcript t, shifting RNA sequence reads of a given length one base at a time along the haplotype sequence, and then performing a weighted sum (weighted again based on insert length and the amount of sequencing error) of the number of RNA sequence reads that would match transcript t and exhibit transcript group g and a specific mutation pattern m (which can be normalized again by dividing by the weighted number of RNA reads that can arise from haplotype h and transcript t, regardless of the mutation pattern or transcript group).

[0128] In the example shown in Figure 6A, haplotype H1 is associated with two transcripts, T1 and T2. These two transcript groups give rise to three transcript groups, g1, g2, and g3, as described above with reference to Figure 1E. Therefore, for a combination of H1 and T1, the system can determine the probability that the hypothetical RNA sequence from that H1 exhibits a given unique mutation pattern and combination with one of the transcript groups. In the example shown in Figure 6A, the system can determine the following:

[0129] Regarding the combination of H1 and T1, ●The probability that the hypothetical RNA sequences from H1 and T1 show a combination of the mutation pattern "0,0,1,?" and transcript group g1 is 602A. ●The probability that the hypothetical RNA sequences from H1 and T1 show a combination of the mutation pattern "0,0,?,?" and transcript group g1 is 602B. ●The probability that the hypothetical RNA sequences from H1 and T1 show the mutation pattern "?,?,?,?" and the transcript group g1 is 602C. ●The probability that the hypothetical RNA sequences from H1 and T1 show the mutation pattern "?,?,?,?" and the transcript group g2 is 602D. ● Any other probability (not shown) that the hypothetical RNA sequences from H1 and T1 show a specific mutation pattern and one of the transcript groups g1, g2, and g3.

[0130] Regarding the combination of H1 and T2, ●The probability that the hypothetical RNA sequences from H1 and T2 show the mutation pattern "?,?,?,?" and the transcript group g2 is 602E. ● Any other probability (not shown) that the hypothetical RNA sequences from H1 and T2 show a specific mutation pattern and one of the transcript groups g1, g2, and g3.

[0131] As shown in Figure 6A, each of probabilities 602A-E can be determined by determining how many hypothetical sequence reads could have such order, position, and / or length to capture such an allele pattern. The example in Figure 6A shows this probability 602B as a relatively high probability event by indicating that there are eight hypothetical sequence reads that have such order, position, and / or length to capture such an allele pattern. In contrast, probability 602E is a relatively low probability event because there is only one hypothetical sequence read that has such order, position, and / or length to capture such an allele pattern.

[0132] In other words, for every haplotype transcript (ht), and for every combination (gm) of transcript group and mutation pattern, the system can count the number of possible read pairs that could result in such a combination gm. In some embodiments, the system weights the counted read pairs by the probability of insert length and / or sequencing error, as described below.

[0133] Figure 6B presents a schematic diagram illustrating the usefulness of determining the incidence of haplotype transcripts when selecting tumor-associated peptides or proteins for the development of personalized anti-cancer therapies. Each haplotype-transcript combination encodes a different protein. The only protein with high incidence (as shown in the figure by the relative number of aligned RNA sequence reads) is encoded by transcript1-haplotype1 (T1-H1). T2-H1 and T1-H2 have low expression. T2-H2 has moderate expression. If the selection of a target protein is based solely on protein expression, T1-H1 would be the best target. It should also be noted that T1-H2 is not a good target even if it has high overall T1 expression unrelated to haplotype. This explains why determining transcript-specific expression may be inappropriate for selecting proteins (e.g., tumor antigens) for downstream therapeutic development, and why the ability to determine transcript-haplotype-specific expression (based on determining the incidence of transcript haplotypes) is a more useful selection criterion.

[0134] V. Statistical models for phasing analysis

[0135] Formula for performing VA fading analysis In the case of DNA sequence reads, the probability of finding mutation pattern m depends on (1) the incidence rate (h) of each haplotype and (2) the conditional probability of finding mutation pattern m in a given read. This relationship can be expressed as follows:

number

[0136] In the above equation, TIFF2026520346000114.tif5170 is, as mentioned earlier, a haplotype. Location from TIFF2026520346000115.tif4170 It starts with TIFF2026520346000116.tif5170, and the length of the insert is TIFF2026520346000117.tif4170, actual alleles The DNA reads of TIFF2026520346000118.tif3170 show mutation patterns. This is an indicator that shows whether or not TIFF2026520346000119.tif3170 is presented. TIFF2026520346000120.tif6170 represents the incidence of an unknown haplotype to be resolved by statistical model 262, as discussed below. mh (Figure 7) Similar to TIFF2026520346000121.tif8170, the values ​​determined for DNA by the enumeration and pattern estimator 260 in Figure 2B are proportional to the probability that a read from a DNA molecule of haplotype h exhibits mutation pattern m (without a normalization constant). mh This is calculated independently of any actual sequence reads. It depends on the mutation site, the DNA sequencing error model, and the probability model of the insert length in the read.

[0137] As explained above, TIFF2026520346000122.tif5170 is an actual mutation pattern. DNA reads containing TIFF2026520346000123.tif3170 have mutation patterns This is the probability that the sequence will be determined as if it were TIFF2026520346000124.tif3170. TIFF2026520346000125.tif5170 is a DNA read pair whose insert length This is the probability of having TIFF2026520346000126.tif4170. As explained earlier, this parameter is determined by fitting a frequentist negative binomial model to DNA reads from the entire genome.

[0138] In the denominator of the above equation, W mhis summed over all mutant patterns m', while for molecules, W mh is summed over haplotypes h and counts the number of ways to generate DNA sequence reads for a particular mutant pattern m.

[0139] For RNA sequence reads, the probability of observing a combination of mutant pattern m and transcript group g in a given read depends on (1) the incidence rate of each combination of haplotype transcripts, (2) the haplotype transcript conditional probability of observing the combination of mutant pattern m and transcript group g, and (3) the distance between mutations in the transcript space. This relationship can be expressed as follows. [Number] where p gm is the probability of observing an RNA sequence read showing a unique mutant pattern m in transcript group g.

[0140] In this equation, I(h,t,l,j,g,a) indicates whether haplotype h and transcript t can generate a read of insert length l in transcript group g using read 1 (i.e., the leftmost read of the read pair) and start j (the genomic coordinate of the start of the leftmost read of the read pair), and also using the actual mutant pattern a (in contrast to the observed mutant pattern m).

[0141] TIFF2026520346000128.tif5170 indicates the probability of mis - calling a base or set of bases within an RNA read having the actual mutant pattern a as the observed mutant pattern m (due to sequencing errors). In other words, the probability that a hypothesized RNA sequence read from that haplotype shows a unique mutant pattern in the combination of haplotype and transcript group can be obtained based on the probability of mis - calling tumor RNA sequence reads having unique mutant patterns. In some embodiments, δ amThis can be estimated by simultaneously examining all reads to determine the proportion of single-base mismatches of allele a to allele m that occur at locations in the aligned reads that have not yet been called mutations. In some embodiments, it is possible to assume equal error probabilities regardless of the error class. A similar approach can be taken to model DNA sequencing errors.

[0142] TIFF2026520346000129.tif5170 shows the probability of insert length l (e.g., the distance between the start of read 1 and the end of read 2 for paired-end sequence reads). In other words, the probability that a hypothetical RNA sequence read from that haplotype exhibits a unique mutation pattern in the combination of haplotype and transcript group can be further based on the probability of insert length of tumor RNA sequence reads that have a unique mutation pattern. In some embodiments, for insert length modeling of RNA sequence reads, very long exons are identified, and the recognized insert lengths from sequence reads from those very long exons are fitted to a negative binomial model. In some embodiments, for insert length modeling of DNA sequence reads, all recognized insert lengths are fitted to a negative binomial model. P(l) is then calculated using the fitted negative binomial.

[0143] In the above formula, TIFF2026520346000130.tif5170 represents the unknown incidence of RNA molecules of haplotype h and transcript t, which should be solved by statistical model 262, as discussed below. gmht (As shown in Figure 2B, the values ​​determined for RNA by the enumeration and pattern probability estimator 260 are shown in Figure 7.) (As shown in TIFF2026520346000131.tif8170) is, as described above, a denormalized probability given haplotype h and transcript t that shows a combination of mutation pattern m and transcript group g in a given sequence read determined from RNA sequence read data. gmht This is calculated independently of any actual sequence reads. It relies on the mutation site, the exons defined by the GTF file, the RNA sequencing error model, and the probabilistic model of the read insert length.

[0144] W g’m’ht This is the unnormalized conditional probability of a haplotype-transcript accepting a combination of mutation pattern m' and group g', summed up over all possible combinations of mutation pattern m' and group g'.

[0145] p is the probability of finding a DNA sequence read that exhibits a unique mutation pattern m. m Determining this means that only one transcript and one group of transcripts are effectively present. gm It can be considered a special case for determining this, and therefore it is not necessary to exponentially represent it over g or t. The sequencing error and insert length distribution of RNA reads versus DNA reads may differ.

[0146] Graph display of statistical models for performing VB mutation phasing.

[0147] Figure 7 shows an exemplary statistical model 262 for estimating, according to embodiments of the present disclosure, the set of probabilities that each of several haplotypes exists and gives rise to a set of sequence reads (also called haplotype existence probabilities) as well as a set of haplotype incidence and haplotype transcript incidence. Specifically, according to embodiments of the present disclosure, the statistical model 262 is used to estimate, based on the output of a read pattern counter 258 (i.e., counts of different mutation patterns, counts of different combinations of mutation patterns and transcript groups) and the output of an enumeration and pattern probability estimator 260 (i.e., the various conditional probabilities described above), the set of probabilities (also called haplotype existence probabilities) that each of several haplotypes exists and gives rise to a set of recognized sequence reads (e.g., tumor DNA sequence reads, tumor RNA sequence reads and normal DNA sequence reads) as well as a set of haplotype incidence and haplotype transcript incidence. In other words, the system uses a read pattern counter 258 to count the occurrences of variant patterns and / or variant pattern-transcript combinations, uses an enumeration and pattern probability estimator 260 to determine the conditional probabilities of these variant patterns or variant pattern-transcript combinations under different haplotype situations, and uses a statistical model 262 to estimate the most likely situation, as described below.

[0148] In some embodiments, the statistical model (or probabilistic graphical model) 262 may include a haplotype generation model, a DNA haplotype incidence model (e.g., a DNA Dirichlet-Multinomial model 704, as shown in the non-limiting example shown in Figure 7), and / or a haplotype transcript group model (e.g., an RNA Dirichlet-Multinomial model 706, as shown in the non-limiting example shown in Figure 7). The haplotype generation model determines the probability of haplotype presence in a given sample using the probability of unique mutation patterns determined from sequence read count data. The DNA haplotype incidence model estimates the haplotype incidence in the sample using the set of haplotype presence probabilities determined by the haplotype generation model and the conditional probability that hypothetical DNA sequence reads from a given haplotype exhibit a particular unique mutation pattern. The haplotype transcript group model estimates the incidence of haplotype transcript groups in a sample using a set of haplotype probabilities determined by a haplotype generation model and a conditional probability that hypothetical RNA sequence reads from a given haplotype and transcript group exhibit a specific, unique mutation pattern.

[0149] In some embodiments, for example, the statistical model 262 may include a haplotype generation model 702, a DNA Dirichlet-Multinomial model 704, and / or an RNA Dirichlet-Multinomial model 706. In certain embodiments, the haplotype generation model 702, the DNA Dirichlet-Multinomial model 704, and the RNA Dirichlet-Multinomial model 706 may each include one or more directed acyclic graphs (DAGs) which are further interconnected by one or more directed links and may include each node 708, 710, 712, 714, 716, 718, 720, 722, 724, and 726 associated with one or more discrete values ​​estimated from a range of probability distributions associated with each of the respective nodes 708, 710, 712, 714, 716, 718, 720, 722, 724, and 726.

[0150] Haplotype generation model 702 uses normal haplotype nodes 708 (e.g., N h ), tumor-specific haplotype node 710 (e.g., X h ), and tumor sample haplotype node 712 (e.g., H h ) may include. H of tumor sample haplotype node 712 h This is an indicator variable that shows whether or not haplotype h is present in the sample. The system is H h By estimating this, we can estimate the set of haplotype probabilities. As will be discussed later, by fitting the model, the system can be estimated to H h We can estimate the posterior distribution of (i.e., the distribution of whether or not haplotype h is present in the sample).

[0151] A normal haplotype node 708 (for example, N h) may include an estimated or approximated binary (e.g., a posterior distribution estimated or approximated based on the range of probability distributions associated with a normal haplotype node 708), and if a given normal haplotype is present in one or more samples in question, N h = "1", and if no given normal haplotype exists, N h = "0". N h This can be taken from one or more samples of the subject. In some embodiments, N h This can be estimated using other phasing methods. In some embodiments, haplotypes that do not contain somatic mutations found in normal tissue are also likely to be found in tumor tissue due to the inevitable contamination of tumor samples by normal cells. In some embodiments, this can be explained by giving a very high prior probability to such haplotypes. In some examples, germline mutant phasing is likely to be conserved in tumor haplotypes that further contain some somatic mutations. If this is true, a higher prior probability can be given to tumor haplotypes that preserve germline mutant structures.

[0152] Tumor-specific haplotype node 710 (e.g., X h ) may include an estimated or approximated binary (e.g., a posterior distribution estimated or approximated based on the range of probability distributions associated with tumor-specific haplotype node 710), and if a given tumor-specific haplotype is present in one or more samples of the subject, X h = "1", and if the given tumor-specific haplotype is not present in one or more samples, X h = "0".

[0153] In certain embodiments, a normal haplotype node 708 (for example, N h ) and tumor-specific haplotype node 710 (e.g., X h ) is connected to each directed link by tumor sample haplotype node 712 (e.g., H h ) can be connected to H hThis includes normal haplogroup Nh and tumor haplogroup X. h It is a mixture of the above. Therefore, in one embodiment, the tumor sample haplotype node 712 (for example, H h ) is an estimate or approximate binary value (e.g., value N h and X h It may include (which is estimated or approximated based on the sum of N) h = "1" or X h = If "1", H h = "1", indicating the presence of at least one normal haplotype or tumor-specific haplotype in the bulk tumor sample. In certain embodiments, one or more hyperprias have a value of "π" (X h It can be represented by (connected to), which can be used to incorporate knowledge of arbitrary prior probabilities of bulk tumor samples into the haplotype generation model 702. In some embodiments, for example, a weak ultraprior distribution for Bernoulli variables can be used. As discussed below, the system can be used ultraprior before sampling tumor haplotypes (and optionally normal haplotypes). Normal haplotype phasing can be used to adjust the prior / ultraprior probabilities regarding the presence of tumor haplotypes. In some embodiments, normal haplotypes are determined based on the use of preprocessing tools (e.g., HapCUT2, WhatsHap). As another alternative, the system can be N h and X h Possible locations of origin can be mixed and matched. For example, a normal haplotype can be identified as a known quantity measured by some other method, such as long reads, microarray data, mother-father-child 3-person information, dilution-pool sequencing, or single-cell sequencing.

[0154] Tumor sample haplotype node 712 (e.g., H h ) is the DNA Dirichlet-Multinomial model 704 haplotype incidence node 714 (for example) It is constrained by TIFF2026520346000132.tif6170) and can be connected by a directed link. In other words, H h This determines whether the occurrence rate node is non-zero.

[0155] Haplotype incidence rate node 714 TIFF2026520346000133.tif6170 is the percentage of DNA molecules in tumor samples derived from haplotype h. The system is, By estimating TIFF2026520346000134.tif6170, the set of haplotype incidence rates can be estimated. As will be described later, by fitting the model, the system The posterior distribution of TIFF2026520346000135.tif6170 can be estimated. As discussed above and shown in the DNA Dirichlet-Multinomial model 704, the haplotype incidence node 714 TIFF2026520346000136.tif6170 can be incorporated into the model and connected to actual sequence reads recognized via the following formula:

number

[0156] TIFF2026520346000138.tif8170 (for example, as represented by parameter 721) represents the denormalized probabilities that, given that the DNA molecule is haplotype h, a hypothetical DNA sequence read may exhibit a unique mutation pattern m, which can be provided by the enumeration and pattern probability estimator 260.

[0157] TIFF2026520346000139.tif4170 (for example, as represented by the probability node parameter 718) represents the probability of observing a unique mutation pattern m. The probability node parameter 718 is the DNA pattern counter node 720 (for example It directs TIFF2026520346000140.tif5170), and is therefore connected by a directed link, and its value TIFF2026520346000141.tif5170 can represent observed DNA pattern counts present in one or more samples, which may be provided by the reading pattern counter 258.

[0158] RNA mutation patterns are modeled similarly to DNA patterns, but the system also explains different transcripts within the same gene. Regarding Figure 7, tumor sample haplotype node 712 (e.g., H h ) is the RNA Dirichlet-Multinomial model 706 haplotype-transcript generation node 722 (for example) It is further constrained by TIFF2026520346000142.tif5170) and can be connected by a directed link. In other words, H h This determines whether the occurrence rate node is non-zero.

[0159] Haplotype transcript occurrence rate node 722 TIFF2026520346000143.tif5170 represents the percentage of RNA molecules in tumor samples derived from haplotype h and transcript t. The system is, By estimating TIFF2026520346000144.tif5170, the set of haplotype transcript occurrence rates can be estimated. As will be described later, by fitting the model, the system The posterior distribution of TIFF2026520346000145.tif5170 can be estimated. As mentioned above, and as shown in the DNA Dirichlet-Multinomial model 704, the estimate or approximation TIFF2026520346000146.tif5170 can be estimated or approximated based on the following formula.

[0160] p(reads have mutation pattern m and transcript group g) =

number

[0161] TIFF2026520346000148.tif8170 (for example, as represented by parameter 727) can show that, given that the RNA molecule is haplotype h and originates from transcript t, the hypothetical RNA sequence reads from the molecule possess a unique mutation pattern m and properties that can be determined by transcript group g-enumeration and pattern probability estimator 210.

[0162] TIFF2026520346000149.tif4170 (for example, as represented by parameter 724) represents the probability of recognizing reads that show a unique combination of mutation pattern m and transcript group g. The parameter is RNA pattern counter node 726 (for example, Instructed to TIFF2026520346000150.tif5170), and thus connected by a directional link, The value of TIFF2026520346000151.tif5170 may represent the recognized count of mutation pattern-transcript group combinations present in one or more samples, which may be provided by the read pattern counter 258. The probability of recognizing a read pair with a given mutation and transcript group pattern may depend on the length of the nucleic acid fragment being sequenced (i.e., the "insert size"). Thus, the system considers a given pattern (p in the above equation). gmTo determine the probability of finding a read pair in ), we estimate the probability of a given fragment length (p(l) in the above equation). To achieve this, the system extracts read pairs that align to long exons (e.g., exons longer than 200bp, 300bp, 400bp, 500bp, 600bp, 700bp, 800bp, 900bp, or 1000bp) that do not overlap with introns from other isoforms. In some embodiments, for example, the system can use the Python package pyranges to identify long exons and the Python package pysam to identify associated read pairs. The absence of introns makes it possible to determine the length of read pair fragments by verifying their initial and final alignment coordinates (i.e., the start of the upstream read and the end of the downstream read). In some embodiments, the system can model the empirical distribution of the lengths of the obtained fragments using a negative binomial distribution that can be fitted using, for example, the Python package statsmodels, the R package MASS, or similar tools that enable fitting of negative binomial distributions to data. The final estimated distribution allows for the determination of p(l) for any given fragment length l.

[0163] The probability of recognizing a read pair with a given mutation pattern may depend on the probability of misidentification of nucleotide bases in the sequencing data. For example, a misidentification of a base may result in a read pair recognizing the mutation pattern "0,1,1" in data from a sample containing only haplotypes "0,0,0" and "1,1,1". Therefore, the system calculates the probability δ that the recognized pattern is different from the actual pattern. am It is necessary to estimate δ (see the formula above). am is, δ am =ε dThe observed pattern can be further decomposed into d, the number of erroneous base calls, and ε, the probability of each individual erroneous base call, such that (assuming all types of miscalls are equally possible). ε can be further estimated directly from the data. In some embodiments, the system can make this estimation by identifying all genomic locations across all sequence reads from a sample lacking the called mutation, and then searching for sequence reads with mutations that were not considered true mutations by the mutant caller used. In some embodiments, the system can query alignments at these non-mutant locations using, for example, the Python package pysam, the C utility samtools, or any other tool that facilitates the manipulation of aligned sequences, and the overall error rate ε can be estimated by the ratio of alignments with "erroneous" (i.e., non-referenced) bases at those locations to all alignments at those locations.

[0164] When using statistical model 262 to evaluate sequence reads derived from a sample from a subject, the system can sample a set of possible haplotypes from a region of the subject's genome. In some embodiments, sampling can be performed on a region from a healthy genome, a region from a cancerous genome resulting from tumor DNA and tumor RNA sequences, or a mixture of (i) a region from a healthy genome and (ii) a region from a cancerous genome resulting from tumor DNA and tumor RNA sequence reads (e.g., sampling of haplotypes present in either a sample from a healthy genome or a sample from a cancerous genome). In some embodiments, for each candidate haplotype in the set of all possible haplotypes, the value of an indicator variable indicating the presence of the haplotype in tumor tissue is sampled from a Bernoulli distribution with a beta prior distribution.

[0165] Referring to Figure 7, when fitting a statistical model (e.g., a stochastic graphical model), the parameter X h(Node 710), H h (Node 712) TIFF2026520346000152.tif6170 (Node 714), TIFF2026520346000153.tif8170 (node ​​718), TIFF2026520346000154.tif6170 (node ​​722), Each of TIFF2026520346000155.tif8170 (node ​​724) can be sampled. Based on the sampling, the system can estimate the posterior distributions of all six of these parameters. Thus, the system can estimate the posterior probability distribution of the incidence of each haplotype, the posterior probability distribution of the incidence of each haplotype transcript, and the posterior probability distribution of the presence of each haplotype in a set of possible haplotypes. In some embodiments, the system uses a Markov chain Monte Carlo (MCMC) sampling method during the fitting process. In some embodiments, the estimated posterior probability distribution of a given parameter is derived from the sampled values ​​of that parameter over multiple samples (e.g., at least 500, 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000, or 10,000 samples). The sampling-based method used in the disclosed method (as opposed to other methods that rely on determining a single-point estimate of the probability) is TIFF2026520346000156.tif6170 and easily provide a measure of uncertainty in the estimates. With this uncertainty at hand, downstream neoantigen selection is estimated TIFF2026520346000157.tif6170 and If the uncertainty surrounding TIFF2026520346000158.tif6170 is high, penalties may be imposed on neoantigens.

[0166] In some embodiments, the statistical model may be a hierarchical Bayesian model. In some embodiments, the Bayesian analysis can be performed using algorithms such as RJAGS or PyMC3. RJAGS provides an interface from R to the JAGS library for Bayesian data analysis.

[0167] VI. Performance Evaluation

[0168] Figure 8A shows the performance evaluation of the techniques described herein on simulated data. Specifically, the performance of only the probabilistic graphical model (PGM) portion of the system (Statistical Model 262 in Figures 2B and 7) is evaluated here by fitting the PGM to simulated read counts and simulated mutation locations, and then comparing the PGM output to simulated truth. This is in contrast to an end-to-end system evaluation on simulated read data, which simulates phasing mutations, then reads data matching those mutations, aligns those simulated reads, then applies the comprehensive system to the aligned reads, and compares the system output to simulated truth.

[0169] The PGM evaluation process is described in more detail below.

[0170] Given a gene whose PGM performance is to be evaluated, the initial generation model randomly samples (i) one of the transcripts of that gene, and then (ii) a mutant location landing in an exon of that transcript; the first mutant location can be anywhere in any exon, and subsequent mutant locations are restricted to being within a certain distance (of bases in transcript space) of any already sampled mutant location. The mutants at these locations themselves are designated as bialleles and are then provided as input to enumerate all possible haplotypes per gene.

[0171] Using the set of mutations and all possible haplotypes at hand, together with a standard GTF specifying the possible transcripts of each gene, a second generative model with a structure almost identical to the PGM was then used to generate the following: (i) a simulated set of existing haplotypes, i.e., a subset of all possible haplotypes; (ii) the incidence rate of each existing haplotype; (iii) the transcript haplotype incidence rate; (iv) the number of DNA reads for each mutation pattern; and (v) the number of RNA reads for each combination of mutation pattern and transcript group.

[0172] There are two essential differences between this second generative model and the actual PGM262. The following exist: (i) Instead of using a beta-binomial, the process by which existing tumor haplotypes are sampled, the P of the existing possible enumerated haplotypes is categorically sampled, where P is a user-specified integer input, and (ii) the total number of reads is not from the bam file but is based on a user-input average read coverage parameter.

[0173] Once the simulated truth data is generated, the actual PGM (RNA-only mode, DNA-only mode, or composite mode) is applied to the simulated read counts to estimate the existing haplotypes, their incidence rates, and the haplotype-transcript incidence rates. The specific point estimates (using the mean) of these parameters are compared to the ground truth. The process of end-to-end comparing the simulated truth data and its estimates is repeated over many iterations of the simulation.

[0174] Specifically, for Figures 8A and 8B, the simulated data parameters were the following: 4 mutations, 100 read lengths, maximum distance between 100 mutations, 200x read range, P=4 haplotypes, and a sequencing error probability of 0.01. Seven genes were evaluated: SNN with 1 transcript, DEFB125 with 2, MAP2K1 with 3, SMYD1 with 4, IAPP with 5, PPP4R4 with 6, and MCPH1 with 7. The default number of repeats was 50.

[0175] In Figure 8A, the mean absolute error (MAE) of the haplotype incidence estimates is plotted as a function of sequencing coverage for simulated sets of DNA sequence read data, RNA sequence read data, and combinations of DNA sequence read data and RNA sequence read data, comparing the performance of the models in various modes. As described elsewhere in this specification, the disclosed methods can be performed using DNA sequence read data (e.g., using models 702 and 704 in Figure 7), RNA sequence read data (e.g., using models 702 and 706 in Figure 7), or a combination of both (e.g., using models 702, 704, and 706 in Figure 7). As shown, incorporating both RNA and DNA data reduces the mean absolute error (MAE) in the haplotype abundance estimate. Furthermore, as sequencing coverage (mean coverage considering gene length) increases, the MAE decreases, leading to a more accurate estimate of haplotype abundance. More broadly, this figure also shows that this model can, in principle, converge to a very accurate solution regarding the incidence of haplotypes, and is not over-parameterized given the expected amount of actual data.

[0176] The data plotted in Figure 8B are similar to those shown in Figure 8A, but the difference is that this figure shows performance based on the mean absolute error (MAE) of the haplotype transcript abundance estimates. As shown in this figure, incorporating both RNA and DNA data reduces the mean absolute error (MAE) of the haplotype transcript abundance estimates. Furthermore, as sequencing coverage (mean coverage considering gene length) increases, the MAE decreases, leading to more accurate estimates of haplotype transcript abundances. More broadly, this figure also shows that this model can, in principle, converge to a very accurate solution for haplotype transcript incidence and is not over-parameterized given the expected amount of read data.

[0177] Figure 8C shows the ground truth φ in the genomic DNA of NA12878, a diploid cell line with a previously haplotype-degraded genome. h φ demonstrates a close agreement with the value. hThe performance evaluation data for the estimated values ​​is shown below. NA12878 sequencing data was downloaded from the NIH Sequence Read Archive (SRA): specifically, i) DNA normal tissue sequencing data with ID SRR10134980 generated by Illumina NovaSeq 6000 using whole exome sequencing, including 242.3 million paired end reads, each 150 bp long; and ii) RNA normal tissue sequencing data with ID SRR19762225 generated by Illumina NovaSeq 6000 using Poly-A mRNA sequencing, including 32 million paired end reads, each 150 bp long. The gold standard haplotype-degrading variant calls were downloaded from Illumina Platinum Genomes (the algorithms related to these calls are described in Eberle, MA et al. (2017), “A reference data set of 5.4 million phased human variants validated by genetic inheritance from sequencing a three-generation 17-member pedigree”, Genome Research 27:157-164).

[0178] With normal (WES)DNA sequencing and RNA sequencing available, and along with gold-standard phased mutant calls, the inventors analyzed the data as follows: WES data were aligned to GRCh38 (Ensembl 100) to generate a beam using BWA-MEM2. RNA-seq data were also aligned to GRCh38 using STAR-2.7.9a. Transcript expression was quantified using Salmon 1.3.0. Next, normal DNA bam was treated as if it were derived from tumor DNA, and gold standard germline variant calls were treated as if they were somatic mutations, and the inventors then used Hh and To predict TIFF2026520346000159.tif6170, a statistical model (as described in Figure 7) was applied, specifically in its DNA-only mode. Fading was limited to genes having at least one TPM estimated by Salmon and at least 10 RNA-seq reads. While this filter was not strictly necessary, in the context of neoantigen characterization, low-expression genes are inadequate targets, and avoiding calculations for them simply reduces runtime in a manner consistent with our intended use. We then calculated the model-estimated haplotype incidence (φ) for NA12878. h (determined as the mean of the posterior probability distribution) versus ground truth φ h The values ​​were compared—this line is the diploid line, and the inventors assume that the ground truth incidence for any haplotype present in the phasing gold standard variant call is 50% or 100%, and for haplotypes missing from the gold standard data, the ground truth incidence is 0%.

[0179] Figure 8D shows performance evaluation data for using the disclosed method to perform binary classification of haplotypes in NA12878 genomic DNA (i.e., the determination of whether a given haplotype is present or not, as estimated by the mean Hh from the modeled sample). The NA12878 genomic DNA is derived from a diploid cell line with known haplotype complements, and therefore known ground truth values ​​for haplotype binary classification. The data for this non-limiting example include all phasing windows containing more than one mutation and six or fewer mutations (i.e., phasing windows containing two, three, four, five, or six mutations; trials also included phasing windows containing eight, nine, or ten mutations). In this receiver operating characteristic (ROC) curve, the true positive rate (TPR) was plotted against the false positive rate (FPR). The true positive rate (TPR = TP / (TP + FN), where TP = #number of true positives and FN = #number of false negatives) and the false positive rate (FPR = FP / (FP + TN), where FP = #number of false positives and TN = #number of true negatives) were calculated for the estimated haplotypes using different H (posterior mean) thresholds. The area under the curve was 0.996.

[0180] Figure 8E shows further performance evaluation data for using the disclosed method to perform binary classification of haplotypes in NA12878 genomic DNA, as described above for Figure 8D. This non-limiting example data includes all phasing windows containing more than one mutation and six or fewer mutations (i.e., phasing windows containing two, three, four, five, or six mutations; trials also included phasing windows containing eight, nine, or ten mutations). Precision was plotted against recall in this receiver operating characteristic (ROC) curve. Precision (=TP / (TP+FP)) and recall (=TPR=TP / (TP+FN)) were calculated with different H (posterior mean) thresholds for the estimated haplotypes. The area under the curve was 0.936.

[0181] Figure 9 shows a flowchart of Method 900 for carrying out one or more computation-based methods for phasing mutations in a tumor of interest, according to the disclosed embodiments. Method 900 may be carried out using one or more processing units (e.g., computing devices and artificial intelligence architectures described below in relation to Figures 10 and 11) which may include hardware (e.g., general-purpose processors, graphics processing units (GPUs), application integrated circuits (ASICs), system-on-a-chip (SoCs), microcontrollers, field-programmable gate arrays (FPGAs), central processing units (CPUs), application processors (APs), visual processing units (VPUs), neural processing units (NPUs), neural decision processors (NDPs), deep learning processors (DLPs), tensor processing units (TPUs), neuromorphic processing units (NPUs), or any other artificial intelligence (AI) / machine learning (ML) accelerator devices that may be suitable for processing medical data and making one or more predictions or decisions thereon), firmware (e.g., microcode), or any combination thereof.

[0182] In some embodiments, method 900 may begin in step 902. In step 902, multiple sequence reads derived from tumor cells obtained from the subject may be accessed, the sequence reads including tumor DNA sequence reads and / or tumor RNA sequence reads. In one or more examples, accessing sequence reads may further include accessing multiple normal DNA sequence reads derived from healthy cells obtained from the subject. In some embodiments, accessing multiple sequence reads may further include accessing a set of germline variant and somatic variant calls derived from tumor cells obtained from the subject.

[0183] In step 904, based on the sequence read data, a set of unique mutation patterns observed in multiple sequence reads can be enumerated.

[0184] In step 906, the set of unique patterns observed in the sequence reads can be counted to calculate the amount of each unique mutation pattern. In one or more examples, counting the set of unique patterns may include calculating the amount of each unique mutation pattern in a normal DNA sequence read. In one or more examples, counting the set of unique patterns may include calculating the amount of each unique mutation pattern in a tumor DNA sequence read, and calculating the amount of each unique mutation and transcript group pattern in a tumor RNA sequence read.

[0185] In step 908, for each unique mutation pattern and for each set of haplotypes, the probability that a hypothetical DNA sequence read from a haplotype can generate a unique mutation pattern can be determined.

[0186] In step 910, for each set of haplotypes and sets of transcripts overlapping with the haplotype, the probability that the hypothetical RNA sequence from that haplotype and its transcripts exhibit a unique mutation pattern in combination with the transcript group can be determined. In one or more examples, the probability that the hypothetical RNA sequence from that haplotype and transcript exhibits a unique combined mutation pattern and transcript group may further depend on the probability of misrecalling a tumor RNA sequence read that has a unique mutation pattern. In one or more examples, the probability that the hypothetical DNA sequence from that haplotype exhibits a unique combined mutation pattern may further depend on the probability of misrecalling a tumor DNA sequence read that has a unique mutation pattern. In one or more examples, the probability that the hypothetical RNA sequence from that haplotype and transcript exhibits a unique mutation pattern and transcript group may further depend on the probability of the insert length of a tumor RNA sequence read that has a unique mutation pattern. In one or more examples, the probability that the hypothetical DNA sequence from that haplotype exhibits a unique mutation pattern may further depend on the probability of the insert length of a tumor DNA sequence read that has a unique mutation pattern.

[0187] In step 912, the amount of mutation patterns and the probabilities of mutation patterns can be input into a statistical model to identify a set of probabilities of each haplotype being present in a sequence read, a set of haplotype incidence rates, and a set of haplotype-transcript incidence rates. In some embodiments, estimating the probability of haplotype presence, the incidence of haplotypes, and the incidence of haplotype-transcripts involves, for each haplotype in the set of candidate haplotypes, using the statistical model to sample the posterior probability distribution for haplotype presence to determine a point estimate (e.g., mean probability), sample the posterior probability distribution for haplotype incidence to determine a point estimate (e.g., mean probability), and sample the posterior probability distribution for haplotype transcript incidence to determine a point estimate (e.g., mean probability) for haplotype transcript incidence. In one or more examples, the set of possible haplotypes may include haplotypes that may be present in a healthy genome for a given region. In one or more examples, the set of possible haplotypes may include haplotypes that may be present in the oncogenic genome for the region. In one or more examples, the set of haplotypes may include haplotypes that may be present in a mixture of (i) the healthy genome for the region and (ii) the oncogenic genome for the region provided by tumor DNA sequence reads and tumor RNA sequence reads. In one or more examples, the set of possible haplotypes that may be present in the oncogenic genome is sampled from a Bernoulli distribution using a beta random variable. In one or more examples, the set of possible haplotypes that may be present in a mixture of (i) the healthy genome and (ii) the oncogenic genome includes haplotypes that are present in the healthy genome, or the oncogenic genome, or both.

[0188] In step 914, the statistical model may output a set of haplotype probabilities, a set of haplotype incidence rates, and a set of haplotype-transcript incidence rates.

[0189] In some embodiments, Method 900 may include identifying a set of peptide sequences using the translation of one or more haplotypes and / or haplotype transcripts based on a set of haplotype probability. In one or more examples, one or more haplotypes and / or haplotype transcripts are associated with non-zero probabilities.

[0190] In some embodiments, Method 900 may further include the step of selecting one or more peptide sequences from a set of peptide sequences based on a set of haplotype incidence rates and / or a set of haplotype transcript incidence rates. In one or more examples, one or more peptide sequences may be further selected based on a predetermined incidence threshold. In one or more examples, one or more peptide sequences may be further selected based on a ranking of the set of peptide sequences based on haplotype incidence rates and / or haplotype transcript incidence rates. In one or more examples, Method 900 may further include the step of identifying a set of mutant peptide sequences based on one or more peptide sequences. In one or more examples, Method 900 may further include the step of generating predictions of the likelihood of presentation of the set of mutant peptide sequences in one or more major histocompatibility complexes (MHCs) or the immunogenicity of one or more of the set of mutant peptide sequences by a machine learning model.

[0191] In some embodiments, Method 900 may further include synthesizing one or more peptides (e.g., using one or more nucleic acid sequences encoding one or more peptides) or precursors into one or more peptides, the one or more peptides including peptides (e.g., mutant peptides or neoantigens) selected based on haplotype and / or haplotype transcript occurrences determined using the Method herein. The synthesized peptides or precursors can then be used experimentally to identify corresponding presentation and / or binding data (e.g., to verify predicted presentation and / or binding, or to generate results to be used for training). For example, the experiment may include evaluating the binding affinity between the selected peptides and specific MHC molecules using an ELISA pull-down assay, a gel shift assay, or a biosensor-based methodology. As another example, the experiment may include collecting elution data indicating whether the selected peptides were presented by MHC molecules by using peptide-MHC immunoprecipitation, followed by detection of the presented MHC ligands by elution and mass spectrometry.

[0192] In addition to validation data indicating whether individual peptides bound to and / or were presented by individual MHCs, or alternatively, validation data may indicate whether individual peptides induced immunogenicity. Immunogenicity results may be determined using in vivo or in vitro tests. Testing one or more selected peptides can be configured to examine one or more immunogenic factors (e.g., to determine whether a given event occurs and / or to what extent) and / or immunogenicity (e.g., to determine whether a peptide induces an immune response and / or to what extent). A test can be configured to examine whether administration of a composition containing one or more peptides (e.g., a vaccine) to a given subject (e.g., the MHC sequences used during mutant peptide selection have been identified) is effective in preventing or treating a medical condition (e.g., a tumor) or disease (e.g., cancer). The subject may be a human subject.

[0193] In some embodiments, method 900 may further include designing and / or manufacturing a pharmaceutical composition (e.g., a personalized cancer immunotherapy and / or a personalized cancer vaccine) based on one or more selected mutant peptides corresponding to all or part of one or more neoantigens (or designing and / or manufacturing a plurality of nucleic acids encoding one or more selected mutant peptides). For example, each of the one or more selected mutant peptides may be predicted to bind to and thereby be presented by the MHC molecules of the subject (e.g., to at least a threshold extent). The pharmaceutical composition may include one or more selected mutant peptides, one or more precursors to the one or more selected mutant peptides, one or more polypeptide sequences corresponding to the one or more selected mutant peptides, RNA (e.g., mRNA) corresponding to the one or more selected mutant peptides, DNA corresponding to the one or more selected mutant peptides, cells (e.g., antigen-presenting cells) containing one or more selected mutant peptides and / or nucleic acids encoding such peptides, plasmids corresponding to the one or more selected mutant peptides, and / or vectors corresponding to the one or more selected mutant peptides.

[0194] The pharmaceutical composition may further include an adjuvant, an excipient, an immunomodulator, a checkpoint protein, an antagonist of PD-1 (e.g., an anti-PD-1 antibody) and / or an antagonist of PD-L1 (e.g., an anti-PD-L1 antibody). The pharmaceutical composition may be a vaccine such as a tumor vaccine. The composition may be a personalized vaccine manufactured or selected for a particular subject.

[0195] A pharmaceutical composition may comprise a polynucleotide construct (e.g., a DNA construct or an RNA construct). A polynucleotide construct is an artificially constructed segment of nucleic acid that can be “implanted” into a target tissue or cell. A polynucleotide construct comprises a DNA or RNA (e.g., mRNA) insertion containing a nucleotide sequence encoding one or more selected mutant peptides. To enhance antigen presentation (e.g., presentation of one or more selected mutant peptides by an MHC molecule), the polynucleotide construct may further comprise modifications developed for improved antigen presentation, and therefore improved immunogenicity to one or more selected mutant peptides. In some examples, the modifications are the incorporation of transmembrane and cytoplasmic regions of the MHC molecule chain into the polynucleotide construct, as described in International Publication No. 2005038030A1, which is incorporated herein by reference in whole for all purposes.

[0196] To provide RNA inserts with increased stability and translation efficiency, polynucleotide constructs may further include modifications developed for improved stability and translation, and therefore improved immunogenicity to one or more selected mutant peptides. In some examples, the modification is the incorporation into a polynucleotide construct of a nucleic acid sequence having at least two copies of the 3' untranslated region of the human β-globin gene, as described in International Publication No. 2007036366A2, which is incorporated herein by reference in its entirety for all purposes. In other examples, the modification is the incorporation of a nucleic acid sequence encoding a 3' untranslated region, such as the F1 3'UTR, as described in International Publication No. 2017060314A3, which is incorporated herein by reference in its entirety for all purposes.

[0197] To provide RNA insertions with increased stability and expression, polynucleotide constructs may further include modifications developed for improved stability and expression, and therefore for improved immunogenicity to one or more selected mutant peptides. In some examples, the modification is the incorporation of a cap at the end of the RNA (e.g., a 5'-cap structure). The cap structure may be a D1 diastereomer of beta-S-ARCA, as described in International Publication No. 2011015347A1, which is incorporated herein by reference in whole for all purposes.

[0198] To deliver polynucleotide constructs to antigen-presenting cells with high selectivity, the composition may further comprise cationic liposomes or lipoplexes to improve the uptake of the polynucleotide constructs and thus improve immunogenicity to one or more selected mutant peptides. In some examples, the composition comprises nanoparticles containing the polynucleotide constructs. The nanoparticles may be lipoplexes comprising one or more lipids such as DOTMA and DOPE, as described in International Publication No. 2013143683A1, which is incorporated herein by reference in whole for any purpose.

[0199] In some embodiments, as described above, Method 900 may further include designing an immunotherapy and / or vaccine (e.g., a personalized cancer immunotherapy and / or a personalized anti-cancer vaccine), the immunotherapy and / or vaccine comprising one or more peptide sequences (e.g., neoantigen sequences) selected based on haplotype incidence and / or haplotype transcript incidence determined using the Method herein.

[0200] In some embodiments, as described above, Method 900 may further include manufacturing an immunotherapy and / or vaccine (e.g., a personalized cancer immunotherapy and / or a personalized anti-cancer vaccine), the immunotherapy and / or vaccine comprising one or more peptide sequences (e.g., neoantigen sequences) selected based on haplotype incidence and / or haplotype transcript incidence determined using the Method herein.

[0201] Some embodiments of the disclosed methods further include treating a medical condition (e.g., a tumor) or disease (e.g., cancer) in an individual by administering to the individual an effective amount of a pharmaceutical composition (e.g., an immunotherapy or vaccine) comprising one or more selected mutant peptides, wherein the one or more mutant peptides are selected using the methods disclosed herein. The individual may be the same individual from which the disease sample was collected. In some examples, the vaccine is administered to a different individual compared to the individual from which the disease sample was collected. The different individual may, for example, be related to the individual from which the disease sample was collected, have a genetic risk of developing a particular type of cancer, and / or have an MHC molecule having one, more or all alleles corresponding to the same (or similar) sequence of one or more MHC alleles as the subject from which the disease sample was collected.

[0202] Figure 10 shows an example of one or more computing devices 1000 that can be used to perform the techniques described herein according to embodiments of the present disclosure. In certain embodiments, one or more computing devices 1000 may perform one or more steps of one or more methods described or shown herein. In certain embodiments, one or more computing devices 1000 may provide functionality described or shown herein. In certain embodiments, software running on one or more computing devices 1000 may perform one or more steps of one or more methods described or shown herein, or provide functionality described or shown herein. Certain embodiments include one or more parts of one or more computing devices 1000.

[0203] This disclosure envisions any preferred number of computing systems 1000. This disclosure envisions one or more computing devices 1000 in any preferred physical form. One or more computing devices 1000 could be, but not limited to, an embedded computer system, a system on a chip (SOC), a single-board computer system (SBC) (e.g., a computer on a module (COM) or system on a module (SOM)), a desktop computer system, a laptop or notebook computer system, an interactive kiosk, a mainframe, a mesh of computer systems, a mobile phone, a personal digital assistant (PDA), a server, a tablet computer system, an augmented / virtual reality device, or two or more of these. Where appropriate, one or more computing devices 1000 could be single or distributed, spanning multiple locations, multiple machines, multiple data centers, or residing in a cloud that may include components of one or more clouds of one or more networks.

[0204] Where appropriate, one or more computing devices 1000 may perform one or more steps of one or more methods described or illustrated herein without substantial spatial or temporal limitations. For example, but not limited to, one or more computing devices 1000 may perform one or more steps of one or more methods described or illustrated herein in real time or in batch mode. Where appropriate, one or more computing devices 1000 may perform one or more steps of one or more methods described or illustrated herein at different times or in different locations.

[0205] In certain embodiments, one or more computing devices 1000 include a processor 1002, memory 1004, a database 1006, an input / output (I / O) interface 1008, a communication interface 1010, and a bus 1012. While this disclosure describes and illustrates a particular computer system having a particular number of particular components in a particular arrangement, this disclosure contemplates any preferred computer system having any preferred number of any preferred components in any preferred arrangement. In certain embodiments, the processor 1002 includes hardware for executing instructions, such as instructions that make up a computer program. To execute instructions, but not as an limitation, the processor 1002 may retrieve (or fetch) instructions from internal registers, internal caches, memory 1004, or database 1006, decode and execute those instructions, and then write one or more results to internal registers, internal caches, memory 1004, or database 1006. In certain embodiments, the processor 1002 may include one or more internal caches for data, instructions, or addresses. This disclosure envisions a processor 1002 that, where appropriate, includes any preferred number of any preferred internal caches. As an example, but not as an limitation, the processor 1002 may include one or more instruction caches, one or more data caches, and one or more translation lookaside buffers (TLBs). Instructions in the instruction caches may be copies of instructions in memory 1004 or database 1006, and the instruction caches may speed up the retrieval of those instructions by the processor 1002.

[0206] The data in the data cache may be a copy of data in memory 1004 or database 1006 that is the target of an instruction currently being executed by processor 1002, the result of a previous instruction executed by processor 1002 that is accessed by a subsequent instruction executed by processor 1002 or written to memory 1004 or database 1006, or other suitable data. The data cache can speed up read or write operations by processor 1002. The TLB can speed up virtual address translation for processor 1002. In certain embodiments, processor 1002 may include one or more internal registers for data, instructions, or addresses. This disclosure intends for a processor 1002 that includes any suitable number of suitable internal registers, where appropriate. Where appropriate, processor 1002 may include one or more arithmetic logic units (ALUs), may be a multicore processor, or may include one or more processors 1002. This disclosure describes and illustrates a particular processor, but this disclosure intends for any suitable processor.

[0207] In certain embodiments, memory 1004 includes main memory for storing instructions for what the processor 1002 is to execute, or data on which the processor 1002 operates. As an example, but not an limitation, one or more computing devices 1000 may load instructions into memory 1004 from a database 1006 or another source (such as another one or more computing devices 1000). The processor 1002 may then load instructions from memory 1004 into internal registers or an internal cache. To execute the instructions, the processor 1002 may retrieve the instructions from the internal registers or internal cache and decode them. During or after the execution of the instructions, the processor 1002 may write one or more results (which may be intermediate or final results) to the internal registers or internal cache. The processor 1002 may then write one or more of those results to memory 1004.

[0208] In certain embodiments, the processor 1002 executes only instructions in one or more internal registers or internal caches or in memory 1004 (as opposed to the database 1006 or other locations), and operates only on data in one or more internal registers or internal caches or in memory 1004 (as opposed to the database 1006 or other locations). One or more memory buses (each possibly including an address bus and a data bus) may connect the processor 1002 to memory 1004. Bus 1012 may include one or more memory buses, as described below. In certain embodiments, one or more memory management units (MMUs) reside between the processor 1002 and memory 1004 to facilitate access to memory 1004 requested by the processor 1002. In certain embodiments, memory 1004 includes random access memory (RAM). This RAM may be volatile memory, where appropriate. This RAM may be dynamic RAM (DRAM) or static RAM (SRAM), where appropriate. Furthermore, this RAM may be single-port RAM or multi-port RAM, where appropriate. This disclosure intends any suitable RAM. Memory 1004 may include one or more devices 1004, where appropriate. While this disclosure describes and illustrates specific memory, this disclosure assumes any suitable memory.

[0209] In certain embodiments, the database 1006 includes mass storage for data or instructions. As an example, but not an limitation, the database 1006 may include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. The database 1006 may include removable or non-removable (or fixed) media, where appropriate. The database 1006 may be located inside or outside one or more computing devices 1000, where appropriate. In certain embodiments, the database 1006 is non-volatile solid-state memory. In certain embodiments, the database 1006 includes read-only memory (ROM). Where appropriate, this ROM may be a mask program ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically modifiable ROM (EAROM), or flash memory, or a combination of two or more of these. This disclosure envisions a mass database 1006 in any preferred physical form. The database 1006 may, where appropriate, include one or more storage control units that facilitate communication between the processor 1002 and the database 1006. Where appropriate, the database 1006 may include one or more databases 1006. While this disclosure describes and illustrates specific storage devices, this disclosure envisions any preferred storage device.

[0210] In certain embodiments, the I / O interface 1008 includes hardware, software, or both that provide one or more interfaces for communication between one or more computing devices 1000 and one or more I / O devices. The one or more computing devices 1000 may, where appropriate, include one or more of these I / O devices. One or more of these I / O devices may enable communication between a person and one or more computing devices 1000. As an example, but not an limitation, the I / O devices may include a keyboard, keypad, microphone, monitor, mouse, printer, scanner, speaker, still camera, stylus, tablet, touchscreen, trackball, video camera, another suitable I / O device, or two or more combinations thereof. The I / O devices may include one or more sensors. This disclosure contemplates any suitable I / O devices and any suitable I / O interface 1008 for them. Where appropriate, the I / O interface 1008 may include one or more device or software drivers that enable the processor 1002 to drive one or more of these I / O devices. The I / O interface 1008 may include one or more I / O interfaces 1008, where appropriate. While this disclosure describes and illustrates specific I / O interfaces, this disclosure intends to include any suitable I / O interface.

[0211] In certain embodiments, the communication interface 1010 includes hardware, software, or both that provide one or more interfaces for communication (e.g., packet-based communication) between one or more computing devices 1000 and one or more other computing devices 1000 or one or more networks. As an example, but not an limitation, the communication interface 1010 may include a network interface controller (NIC) or network adapter for communicating with Ethernet or other wire-based networks, or a wireless NIC (WNIC) or wireless adapter for communicating with wireless networks such as Wi-Fi networks. This disclosure envisions any suitable network and any suitable communication interface 1010 for that network.

[0212] As an example, and not an limitation, one or more computing devices 1000 may communicate with one or more ad hoc networks, personal area networks (PANs), local area networks (LANs), wide area networks (WANs), metropolitan area networks (MANs), the Internet, or two or more combinations thereof. One or more of these networks may be wired or wireless. For example, one or more computing devices 1000 may communicate with wireless PANs (WPANs) (such as Bluetooth WPANs), Wi-Fi networks, Wi-MAX networks, cellular telephone networks (such as Global System for Mobile Communications (GSM) networks), other suitable wireless networks, or two or more combinations thereof. One or more computing devices 1000 may, where appropriate, include any suitable communication interface 1010 for any of these networks. The communication interface 1010 may, where appropriate, include one or more communication interfaces 1010. While this disclosure describes and illustrates specific communication interfaces, this disclosure intends to include any suitable communication interfaces.

[0213] In certain embodiments, bus 1012 includes hardware, software, or both that connect components of one or more computing devices 1000 to one another. As an example, but not an limitation, bus 1012 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Frontside Bus (FSB), a Hypertransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI Express (PCIe) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or another suitable bus, or two or more combinations thereof. Bus 1012 may include one or more buses 1012, where appropriate. While this disclosure describes and illustrates specific buses, this disclosure intends to cover any suitable bus or interconnect.

[0214] In this specification, computer-readable non-temporary storage media may, where appropriate, include one or more semiconductor-based or other integrated circuits (ICs) (such as field-programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs)), hard disk drives (HDDs), hybrid hard drives (HHDs), optical disks, optical disk drives (ODDs), magneto-optical disks, magneto-optical drives, floppy diskettes, floppy disk drives (FDDs), magnetic tapes, solid-state drives (SSDs), RAM drives, secure digital cards or drives, any other suitable computer-readable non-temporary storage media, or any two or more suitable combinations thereof. Computer-readable non-temporary storage media may, as necessary, be volatile, non-volatile, or a combination of volatile and non-volatile.

[0215] In this specification, “or” is inclusive and not exclusive, unless otherwise explicitly stated or indicated by context. Thus, in this specification, “A or B” means “A, B, or both” unless otherwise explicitly stated or indicated by context. Furthermore, “and” means both jointly and individually unless otherwise explicitly stated or indicated by context. Thus, in this specification, “A and B” means “A and B, jointly or individually” unless otherwise explicitly stated or indicated by context.

[0216] In this specification, “automatically” and its derivatives mean “without human intervention” unless otherwise explicitly stated or indicated by the context.

[0217] The embodiments disclosed herein are illustrative and the scope of this disclosure is not limited thereto. The embodiments provided herein are disclosed in particular in the appendix claims relating to methods, storage media, systems, and computer program products, and any feature mentioned in one claim category, e.g., Method, may also be claimed in another claim category, e.g., System. Dependencies or references in the appendix claims are selected for formal reasons only. However, any subject matter arising from any intentional reference (in particular, multiple dependencies) to any prior claim may also be claimed, and therefore any combination of claims and their features is disclosed and may be claimed regardless of the dependencies selected in the appendix claims. Claimable subject matter includes not only combinations of features as described in the appendix claims, but also any other combination of features within the claims, and each feature described in the claims may be combined with any other feature or combination of features within the claims. Furthermore, any embodiment or feature described or illustrated herein may be claimed in a separate claim, and / or in any combination of any embodiment or feature described or illustrated herein, or any feature of the appended claims.

[0218] The scope of this disclosure includes all variations, substitutions, alterations, modifications, and changes to the exemplary embodiments described or illustrated herein, as would be understood by those skilled in the art. The scope of this disclosure is not limited to the exemplary embodiments described or illustrated herein. Furthermore, while this disclosure describes and illustrates each embodiment herein as including a particular component, element, feature, function, operation, or step, any of these embodiments may include any combination or permutation of any component, element, feature, function, operation, or step described or illustrated anywhere in this specification, as would be understood by those skilled in the art. Furthermore, any claim reference to an attachment to an apparatus or system or component of an apparatus or system that is adapted to, arranged to, capable of, configured to, enabled to, operable to, or operable to perform a particular function includes that apparatus, system, or component, to the extent that that apparatus, system, or component is adapted, arranged, capable, configured, enabled to, operable to, or operable in such a way, whether or not that apparatus, system, component or its particular function is activated, turned on, or unlocked. In addition, while this disclosure describes or indicates certain embodiments as providing certain advantages, certain embodiments may provide none of these advantages, some of them, or all of them.

[0219] Embodiment The embodiments provided include the following: 1 A method for phasing germline variants and somatic variants identified in a target tumor, comprising: accessing a plurality of sequence reads derived from tumor cells obtained from the target using one or more computing devices, wherein the sequence reads include tumor DNA sequence reads and / or tumor RNA sequence reads; enumerating a set of unique variant patterns observed in the plurality of sequence reads; counting the number of sequence reads representing each unique variant pattern in the set of unique variant patterns observed in the sequence reads to calculate the amount of each unique variant pattern; and / or counting the number of sequence reads representing each combination of a unique variant pattern from the set of unique variant patterns and a set of transcripts from one or more transcripts associated with a gene to calculate the amount of each combination of a unique variant pattern and a set of transcripts; for each unique variant pattern, for each haplotype in the set of haplotypes, A method comprising: determining the probability that a hypothetical DNA sequence read from a type exhibits a unique mutation pattern, and / or for each combination of haplotypes from a set of haplotypes, transcripts from a set of gene-associated transcripts, and transcript groups from a set of gene-associated transcript groups, the probability that a hypothetical RNA sequence read from a haplotype and transcript exhibits a unique mutation pattern and transcript group; inputting the amount of unique mutation patterns and / or unique mutation pattern transcript groups, as well as the probability of unique mutation patterns and / or unique mutation pattern transcript groups, into a statistical model to estimate at least one of the following: (i) a set of haplotype existence probabilities in which each haplotype exists, (ii) a set of haplotype incidence rates, and (iii) a set of haplotype transcript incidence rates; and outputting at least one of the following: (i) a set of haplotype existence probabilities, (ii) a set of haplotype incidence rates, and (iii) a set of haplotype transcript incidence rates. The method according to Embodiment 1, further comprising identifying a set of peptide sequences by translating one or more haplotypes and / or haplotype transcripts based on a set of haplotype probability. 3. The method according to Embodiment 2, wherein one or more haplotypes and / or haplotype transcripts are associated with a non-zero probability. 4. The method according to either Embodiment 2 or 3, further comprising selecting one or more peptide sequences from a set of peptide sequences based on a set of haplotype incidence rates and / or a set of haplotype transcript incidence rates. 5. The method according to Embodiment 4, further comprising selecting one or more peptide sequences based on a predetermined incidence threshold. 6. The method according to either Embodiment 4 or 5, wherein the selection of one or more peptide sequences is further based on a ranking of the set of peptide sequences based on a set of haplotype incidence rates and / or a set of haplotype transcript incidence rates. 7. The method according to any one of embodiments 4 to 6, further comprising identifying a set of mutant peptide sequences based on one or more peptide sequences. 8. The method according to Embodiment 7, further comprising using a machine learning model to generate predictions of the likelihood of presentation of a set of mutant peptide sequences in one or more major histocompatibility complexes (MHCs) or the immunogenicity of one or more sets of mutant peptide sequences. The method according to any one of Embodiments 1 to 8, further comprising accessing sequence reads to access multiple normal DNA sequence reads derived from healthy cells obtained from the subject. The method according to Embodiment 9, further comprising counting sets of unique patterns observed in multiple sequence reads to calculate the amount of each unique mutation pattern in a normal DNA sequence read. 11 The method according to any one of Embodiments 1 to 10, further comprising counting sets of unique patterns observed in multiple sequence reads, calculating the amount of each unique mutation pattern in tumor DNA sequence reads, and calculating the amount of each unique mutation and transcript group pattern in tumor RNA sequence reads. 12 The method according to any one of Embodiments 1 to 11, wherein the probability that the hypothetical RNA sequence derived from the haplotype exhibits a unique mutation pattern in combination with the haplotype and transcript group is further based on the probability of misrecalling a tumor RNA sequence read having a unique mutation pattern. 13 The method according to any one of Embodiments 1 to 12, wherein the probability that a hypothetical DNA sequence read from that haplotype exhibits a unique mutation pattern is further based on the probability of misrepresenting a tumor DNA sequence read as having a unique mutation pattern. 14 The method according to any one of Embodiments 1 to 13, wherein the probability that a hypothetical RNA sequence from that haplotype exhibits a unique mutation pattern in combination with the haplotype and transcript group is further based on the probability of the insert length of the RNA sequence read having the unique mutation pattern. 15 The method according to any one of Embodiments 1 to 14, wherein the probability that a hypothetical DNA sequence from that haplotype exhibits a unique mutation pattern is further based on the probability of the insert length of the DNA sequence read having a unique mutation pattern. The method according to any one of Embodiments 1 to 15, further comprising: inputting the amount and probability of mutation patterns into a statistical model to estimate a set of haplotype probabilities, a set of haplotype incidence rates, and a set of haplotype transcript incidence rates; sampling a set of possible haplotypes from a region of the genome in question; estimating the posterior distribution of each haplotype incidence rate for the statistical model based on the sampling; estimating the posterior distribution of each haplotype transcript incidence rate for the statistical model based on the sampling; and estimating the posterior distribution of the presence of each haplotype in the set of possible haplotypes for the statistical model based on the sampling. 17 The method according to Embodiment 16, wherein sampling a set of possible haplotypes includes sampling haplotypes that may be present in a healthy genome for a region. 18. The method according to either embodiment 16 or 17, wherein sampling a set of haplotypes includes sampling haplotypes that may be present in the cancer genome for a region. 19 The method according to any one of Embodiments 16 to 18, wherein sampling a set of haplotypes comprises sampling from haplotypes that may be present in a mixture of (i) healthy genomes for regions and (ii) cancer genomes for regions provided by tumor DNA sequence reads and tumor RNA sequence reads. The method according to any one of embodiments 16 to 19, wherein a set of haplotypes sampled from possible haplotypes present in a cancer genome is sampled from a Bernoulli distribution using a beta random variable. 21 The method according to any one of Embodiments 16 to 20, wherein sampling from a haplotype that may be present in a mixture of (i) a healthy genome and (ii) a cancer genome is a method comprising sampling a haplotype that is present in either a sample of a healthy genome or a sample of a cancer genome. 22 The method according to any one of embodiments 1 to 21, further comprising accessing multiple sequence reads to access a set of germline variants and somatic variant calls derived from tumor cells and normal cells obtained from the subject. A system comprising one or more computing devices, each comprising one or more non-temporary computer-readable storage media containing instructions, and one or more processors coupled to the one or more storage media, wherein one or more processors are configured to execute instructions and carry out the method described in any one of embodiments 1 to 22. 24 A non-temporary computer-readable medium containing instructions that cause one or more processors of one or more computing devices to perform the method described in any one of embodiments 1 to 22 when executed by one or more processors of one or more computing devices. 25 A vaccine comprising: one or more peptides; a plurality of nucleic acids encoding one or more peptides; and a plurality of cells expressing one or more peptides, wherein one or more peptides are selected from a set of peptides by carrying out the method described in any one of Embodiments 1 to 22. 26 A method for producing a vaccine, comprising producing a vaccine comprising one or more peptides, a plurality of nucleic acids encoding one or more peptides, or a plurality of cells expressing one or more peptides, wherein the one or more peptides are selected from a set of peptides by carrying out the method of any one of Embodiments 1 to 22. 27 A pharmaceutical composition comprising one or more peptides selected from a set of peptides by carrying out the method described in any one of Embodiments 1 to 22.

Claims

1. A method for phasing mutations identified in a target tumor, comprising one or more computing devices, Accessing a plurality of sequence reads derived from tumor cells obtained from the subject, wherein the sequence reads include tumor DNA sequence reads and / or tumor RNA sequence reads. To enumerate a set of unique mutation patterns observed in the aforementioned multiple sequence reads, Counting the number of sequence reads that exhibit each of the sets of unique mutation patterns observed in the sequence reads, and / or calculating the amount of each of the unique mutation patterns, and / or The number of sequence reads representing each combination of a unique mutation pattern from a set of unique mutation patterns and a set of transcripts from one or more transcripts associated with the gene is counted to calculate the quantity of each combination of unique mutation patterns and transcripts. For each of the aforementioned unique mutation patterns, For each haplotype in the set of haplotypes, the probability that a hypothetical DNA sequence read from the said haplotype exhibits a unique mutation pattern, and / or For each combination of haplotypes from the set of haplotypes, transcripts from the set of transcripts associated with the gene, and transcript groups from the set of transcript groups associated with the gene, the probability that the hypothetical RNA sequence reads from the haplotypes and transcripts exhibit the unique mutation pattern and transcript groups is determined. (i) inputting into a statistical model the amount of the unique mutation pattern and / or the amount of the unique mutation pattern transcripts, as well as the probability of the unique mutation pattern and / or the probability of the unique mutation pattern transcripts, in order to estimate at least one of the following: (i) a set of haplotype probability sets in which each of the haplotypes exists, (ii) a set of haplotype incidence rates, and (iii) a set of haplotype transcript incidence rates, and A method comprising (i) outputting at least one of the sets of haplotype existence probabilities, (ii) the sets of haplotype occurrence rates, and (iii) the sets of haplotype transcript occurrence rates.

2. At least one of the sets of haplotype probability values, haplotype incidence rates, and haplotype transcript incidence rates is, (i) Using the statistical model, for each haplotype in the set, sample from at least one of the following: the posterior probability distribution of haplotype presence, the posterior probability distribution of haplotype incidence, and the posterior probability distribution of haplotype transcript incidence, and (ii) The method according to claim 1, comprising using a sample from each of the aforementioned posterior probability distributions to calculate point estimates for the probability of haplotype presence, the haplotype incidence, and the haplotype transcript incidence for each haplotype in the set.

3. The method according to claim 1 or 2, further comprising identifying a set of mutant peptide sequences using in silico translation of one or more haplotype sequences and / or haplotype transcripts based on the set of haplotype probability, the set of haplotype incidence rates, and / or the set of haplotype transcript incidence rates.

4. The method according to any one of claims 1 to 3, wherein the one or more haplotype sequences and / or haplotype transcripts are associated with the probability of non-zero haplotype presence.

5. The method according to claim 3 or 4, further comprising selecting one or more mutant peptide sequences from the set of mutant peptide sequences using one or more predetermined criteria, including predetermined criteria applicable to the set of haplotype incidence rates and / or the set of haplotype transcript incidence rates.

6. The method according to claim 5, wherein the one or more predetermined criteria include a predetermined haplotype incidence threshold and / or a predetermined haplotype transcript threshold.

7. The method according to claim 3 or 4, further comprising selecting one or more mutant peptide sequences from the set of mutant peptide sequences by ranking the set of peptide sequences based on the set of haplotype incidence rates and / or the set of haplotype transcript incidence rates.

8. The method according to any one of claims 3 to 7, further comprising using a machine learning model to generate predictions of the likelihood of presentation of one or more of the set of mutant peptide sequences in the major histocompatibility complex (MHC) and / or predictions of immunogenicity for one or more of the set of mutant peptide sequences.

9. The method according to any one of claims 1 to 8, further comprising accessing the sequence reads to access a plurality of normal DNA sequence reads derived from healthy cells obtained from the subject.

10. Counting the set of unique mutation patterns observed in the aforementioned plurality of sequence reads The method according to claim 9, further comprising calculating the amount of each unique mutation pattern in the normal DNA sequence reads.

11. Counting the set of unique mutation patterns observed in the aforementioned plurality of sequence reads Calculating the amount of each unique mutation pattern in the tumor DNA sequence reads, and / or The method according to any one of claims 1 to 10, further comprising calculating the amount of each unique mutation pattern and transcript group combination in the tumor RNA sequence reads.

12. The method according to any one of claims 1 to 11, wherein the probability that the hypothetical RNA sequence reads from the haplotype and transcript exhibit the unique mutation pattern and the transcript group is calculated from the haplotype transcript occurrence rate, the conditional probability of recognizing the combination of the unique mutation pattern in the RNA sequence read and the transcript group, and the length of the transcript.

13. The method according to claim 12, further comprising calculating the probability that the hypothetical RNA sequence reads from the haplotype and transcripts exhibit the unique mutation pattern and the transcript group, taking into account the probability of misrecalling tumor RNA sequence reads having the unique mutation pattern.

14. The method according to any one of claims 1 to 13, wherein the probability that the hypothetical DNA sequence read from the haplotype exhibits the unique mutation pattern is calculated from the haplotype incidence and the conditional probability of recognizing the unique mutation pattern in the DNA sequence read.

15. The method of claim 14, further comprising the calculation of the probability that the hypothetical DNA sequence read from the haplotype exhibits the unique mutation pattern, taking into account the probability of misrepresenting a tumor DNA sequence read as the unique mutation pattern.

16. The method according to any one of claims 1 to 15, wherein the calculation of the probability that the hypothetical RNA sequence from the haplotype and transcript exhibits the unique mutation pattern and the transcript group further includes considering the probability that a given insert length of the RNA sequence read exhibits the unique mutation pattern and the transcript group.

17. The method according to any one of claims 1 to 16, wherein the calculation of the probability that the hypothetical DNA sequence from the haplotype exhibits the unique mutation pattern further includes considering the probability that a given insert length of the DNA sequence read exhibits the unique mutation pattern.

18. The method according to any one of claims 1 to 17, further comprising calculating the probability that the hypothetical RNA sequence from the haplotype and transcript exhibits the unique mutation pattern and the transcript group, taking into account the RNA sequencing error probability.

19. The method according to any one of claims 1 to 18, further comprising calculating the probability that the hypothetical DNA sequences from the haplotype and transcripts exhibit the unique mutation pattern and the transcript group, taking into account the probability of DNA sequencing error.

20. An estimate of at least one of the set of haplotype probability of existence, the set of haplotype incidence rates, and the set of haplotype transcript incidence rates is made for each haplotype in the set of possible haplotypes, using the statistical model. To sample the posterior probability distribution of haplotype existence and determine the point estimate of haplotype existence. To sample the posterior probability distribution of haplotype incidence to determine a point estimate of haplotype incidence, and / or The method according to any one of claims 1 to 19, comprising sampling a posterior probability distribution for the occurrence rate of haplotype transcripts and determining a point estimate for the occurrence rate of haplotype transcripts.

21. The method according to claim 20, wherein the point estimate for each posterior probability distribution includes the mean.

22. The method according to claim 20 or 21, wherein the posterior probability distributions for haplotype presence, haplotype incidence, and / or haplotype transcript incidence are further used to determine the uncertainty associated with haplotype presence, haplotype incidence, and / or haplotype transcript incidence, respectively.

23. The method according to any one of claims 1 to 22, further comprising accessing the plurality of sequence reads to access a set of germline variants and somatic variant calls derived from tumor cells obtained from the subject.

24. The method according to any one of claims 1 to 23, wherein the statistical model includes a probabilistic graphical model.

25. The method according to any one of claims 1 to 24, wherein the statistical model includes a hierarchical Bayesian model.

26. The method according to claim 25, wherein the hierarchical Bayesian model includes a haplotype generation model, a DNA Dirichlet-Multinomial model, and / or an RNA Dirichlet-Multinomial model.

27. A system including one or more computing devices, One or more non-temporary computer-readable storage media containing instructions, and One or more processors coupled to one or more of the storage media, Equipped with, A system in which one or more processors are configured to execute instructions for carrying out the method according to any one of claims 1 to 26.

28. A non-temporary computer-readable medium containing instructions that cause one or more processors of one or more computing devices to perform the method according to any one of claims 1 to 26 when executed by said processors.

29. It is a vaccine, One or more peptides, Multiple nucleic acids encoding one or more peptides, or A vaccine comprising a plurality of cells expressing one or more of the peptides, wherein the one or more peptides are selected from a set of peptides by carrying out the method according to any one of claims 1 to 26.

30. A method for manufacturing vaccines, To select one or more peptides from a set of peptides, the method described in any one of claims 1 to 26 is carried out, Including the production of the vaccine, the vaccine One or more selected peptides, at least one of them; Multiple nucleic acids encoding at least one of the selected peptides, or Multiple cells expressing at least one of the one or more peptides mentioned above, Methods that include...

31. A pharmaceutical composition comprising one or more peptides selected from a set of peptides by carrying out the method described in any one of claims 1 to 26.

Images