Methods and compositions for homologous recombination deficiency detection

An ensemble-based system using targeted sequencing and biomarkers addresses the limitations of HRD detection in tumors lacking BRCA1/2 mutations by integrating SNV and CNV metrics, improving specificity and robustness, and predicting treatment response.

WO2026241440A1PCT designated stage Publication Date: 2026-11-19GUARDANT HEALTH INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/US2026/028063
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-13
Filing Date
2026-05-13
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Current methods for determining homologous recombination deficiency (HRD) in cancer tumors, particularly those lacking canonical BRCA1 or BRCA2 mutations, are inadequate due to reliance on mutation status alone, leading to incomplete detection and false positives in low-input sequencing environments like liquid biopsies.

Method used

An ensemble-based system using targeted next-generation sequencing data integrates multiple orthogonal biomarkers, including SNV mutational signatures, copy-number variation, and genomic scarring metrics, with probabilistic classifiers to accurately determine HRD status, adjusting for tumor ploidy and quality control metrics to improve specificity and robustness.

Benefits of technology

The system effectively identifies HRD-positive tumors, expanding therapeutic eligibility beyond mutation-only testing, predicting treatment response, and enabling longitudinal monitoring across diverse tumor types and sequencing conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US2026028063_19112026_PF_FP_ABST
    Figure US2026028063_19112026_PF_FP_ABST
Patent Text Reader

Abstract

Described herein are systems and methods for determining homologous recombination deficiency (HRD) using targeted next-generation sequencing data derived from tissue and liquid biopsy samples. In various embodiments, a probabilistic classifier utilizes single nucleotide variant mutational signatures, copy-number variation features, genomic scarring biomarkers, epigenomic biomarkers, transcriptomic biomarkers, ploidy-adjusted metrics, and quality control parameters to determine genomic instability status. In some embodiments, the systems utilize ensemble classifiers selectively applying CNV-only and CNV+SNV models depending on somatic variant abundance. The disclosed methods improve sensitivity and specificity for HRD detection and enable prediction of responsiveness to DNA damage response-targeted therapies including PARP inhibitors.
Need to check novelty before this filing date? Find Prior Art

Description

Attorney Docket No. GH0286WOMETHODS AND COMPOSITIONS FOR HOMOLOGOUS RECOMBINATION DEFICIENCY DETECTIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Patent ApplicationNo. 63 / 805,200, filed May 13, 2025, which is fully incorporated by reference herein in entirety.BACKGROUND

[0002] Homologous recombination repair (HRR) is a fundamental cellular pathway responsible for the high-fidelity repair of DNA double-stranded breaks and replication-associated DNA damage. Defects in HRR pathways can lead to homologous recombination deficiency (HRD), a molecular phenotype associated with genomic instability, chromosomal rearrangements, allelic imbalance, and characteristic mutational patterns. Tumors exhibiting HRD are frequently more sensitive to DNA damaging therapies and DNA damage response-targeted agents, including platinum-based chemotherapies and poly(ADP-ribose) polymerase (PARP) inhibitors. Clinical studies in ovarian, prostate, pancreatic, and breast cancers have demonstrated that HRD-positive tumors may derive substantial therapeutic benefit from such treatments through mechanisms including synthetic lethality. Accordingly, there remains a need for improved systems and methods capable of accurately determining HRD using targeted next-generation sequencing assays across both tissue and liquid biopsy sample types. There is further need for computational systems capable of integrating multiple orthogonal biomarker classes, including SNV mutational signatures, copy-number features, genomic scarring biomarkers, methylation features, transcriptomic biomarkers, tumor fraction measurements, and quality control metrics. There is additionally a need for robust probabilistic classifiers capable of functioning reliably in low tumor fraction samples, reducing false-positive calls associated with polyploidy and noisy copy-number modeling, and identifying clinically relevant HRD-positive tumors lacking canonical BRCA1 or BRCA2 mutations.

[0003] The present disclosure addresses these and other needs by providing ensemble-based systems and methods for probabilistic determination of HRD and genomic instability status using targeted next-generation sequencing data derived from tissue and liquid biopsy samples.Attorney Docket No. GH0286WO

[0004] Described herein are systems, methods, computer-implemented methods, software platforms, targeted sequencing assays, classifiers, and reagent kits for determining homologous recombination deficiency (HRD) and genomic instability status (GIS) in biological samples obtained from subjects having or suspected of having cancer. The disclosed technologies are particularly useful for identifying tumors exhibiting functional defects in homologous recombination repair pathways, including tumors lacking canonical BRCA1 or BRCA2 mutations but nevertheless exhibiting genomic scarring and mutational characteristics associated with HRD.SUMMARY OF THE INVENTION

[0005] The present disclosure provides systems, methods, computer-implemented methods, software platforms, targeted sequencing assays, classifiers, and reagent kits for determining homologous recombination deficiency (HRD) and genomic instability status (GIS) in biological samples obtained from subjects having or suspected of having cancer. The disclosed technologies are particularly useful for identifying tumors exhibiting functional defects in homologous recombination repair pathways, including tumors lacking canonical BRCA1 or BRCA2 mutations but nevertheless exhibiting genomic scarring and mutational characteristics associated with HRD.

[0006] In various embodiments, the present disclosure provides computer-implemented methods for determining homologous recombination deficiency (HRD) status in a biological sample. The methods may comprise obtaining sequencing data from a targeted nextgeneration sequencing assay, determining one or more single nucleotide variant (SNV) mutations from the sequencing data, generating a probabilistic HRD score using a classifier based on one or more features including SNV mutations, copy-number variation (CNV) features, loss of heterozygosity (LOH), large-scale state transitions (LST), telomeric allelic imbalance (TAI), allelic imbalance, and copy-number burden, and classifying the biological sample as HRD-positive or HRD-negative based on the probabilistic HRD score.

[0007] In certain embodiments, one or more CNV features are determined from the sequencing data. Such CNV features may comprise one or more of loss of heterozygosity, large-scale state transitions, telomeric allelic imbalance, allelic imbalance, and copy-number burden. In some embodiments, one or more mutational signatures comprise SNV mutations and / or COSMIC mutational signatures. In certain embodiments, the COSMIC signatures comprise SBS3. In other embodiments, the COSMIC signatures comprise SBS3, SBS39, SBS6, SBS15, SBS26, or combinations thereof.Attorney Docket No. GH0286WO

[0008] In some embodiments, the classifier comprises a logistic regression model. In other embodiments, the classifier comprises an ensemble classifier. In certain embodiments, the ensemble classifier selectively applies a CNV-only model and / or a combined CNV+SNV model. In some embodiments, model selection depends on the quantity of detected somatic SNVs. In certain embodiments, the CNV+SNV model is utilized when at least 10, 15, 20, 25, or more somatic SNVs are detected, optionally at least 20 somatic SNVs.

[0009] In various embodiments, the disclosed methods further comprise estimating tumor ploidy. In certain embodiments, one or more LST values are adjusted based on tumor ploidy. In some embodiments, the methods further comprise determining one or more CNV quality control metrics. In certain embodiments, the CNV quality control metrics comprise mean copy-number distance. In other embodiments, the CNV quality control metrics comprise mean mutant allele fraction (MAF) distance. In some embodiments, samples exceeding one or more threshold CNV quality control values are classified as not evaluable.

[0010] In certain embodiments, the biological sample comprises a liquid biopsy sample, a tissue biopsy sample, or combinations thereof. In some embodiments, the liquid biopsy sample comprises circulating tumor DNA and / or cell-free DNA. In other embodiments, the biological sample comprises a tissue sample including a formalin-fixed paraffin-embedded tissue sample.

[0011] In various embodiments, the targeted sequencing assay interrogates at least 500 genes. In other embodiments, the targeted sequencing assay interrogates between approximately 600 and 800 genes, optionally at least 700 genes. In certain embodiments, the sequencing data further comprise methylation biomarkers. In other embodiments, the sequencing data further comprise transcriptomic biomarkers. In some embodiments, the classifier further utilizes epigenomic biomarkers and / or transcriptomic biomarkers.

[0012] In certain embodiments, the probabilistic HRD score ranges from 0 to 1. In some embodiments, the biological sample is classified as HRD detected, HRD not detected, or HRD not evaluable.

[0013] In various embodiments, the disclosed methods further comprise selecting or administering a therapy based on the HRD classification. In certain embodiments, the therapy comprises a PARP inhibitor and / or a base excision repair (BER) inhibitor. In some embodiments, the PARP inhibitor comprises olaparib, rucaparib, niraparib, talazoparib, or combinations thereof. In other embodiments, the therapy comprises platinum chemotherapy. In additional embodiments, the therapy comprises an ATR inhibitor.Attorney Docket No. GH0286WO

[0014] In certain embodiments, the disclosed methods predict responsiveness to DNA damage response-targeted therapy. In some embodiments, the methods predict progression-free survival and / or radiographic progression-free survival. In other embodiments, the methods identify HRD-positive tumors lacking pathogenic BRCA1 and / or BRCA2 mutations.

[0015] In some embodiments, HRD-positive training samples comprise samples exhibiting biallelic BRCA1 and / or BRCA2 loss. In other embodiments, HRD-negative training samples comprise samples lacking deleterious homologous recombination repair (HRR) mutations, promoter methylation in HRR genes, loss of heterozygosity in HRR genes, and / or pathogenic germline HRR variants.

[0016] In various embodiments, mutational signatures are determined using normalized mutational count matrices. In certain embodiments, the normalized mutational count matrices comprise 96 trinucleotide mutational contexts. In some embodiments, the methods further comprise generating mutational signature projection scores using reference mutational signatures. In certain embodiments, the projection scores are generated by matrix decomposition methods. In other embodiments, the projection scores are generated using non-negative least squares methods.

[0017] The present disclosure additionally provides systems configured to perform any of the methods disclosed herein. In certain embodiments, the disclosure provides non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform any of the disclosed methods.

[0018] In various embodiments, the disclosure further provides kits for determining homologous recombination deficiency. Such kits may comprise reagents for targeted nextgeneration sequencing and / or software configured to generate a probabilistic HRD score using SNV mutational signatures and CNV features. In certain embodiments, the software applies a ploidy-adjusted LST score. In other embodiments, the software classifies samples as HRD detected, HRD not detected, or HRD not evaluable.

[0019] In additional embodiments, the disclosure provides methods of monitoring HRD status in a subject over time. Such methods may comprise determining a first HRD score from a first biological sample obtained at a first time point and determining a second HRD score from a second biological sample obtained at a second time point. In some embodiments, one biological sample comprises tissue-derived nucleic acids and another biological sample comprises circulating tumor DNA.Attorney Docket No. GH0286WO

[0020] In various embodiments, the disclosed methods utilize targeted next-generation sequencing data derived from tissue samples, liquid biopsy samples, or combinations thereof. The targeted sequencing assays are configured to profile a plurality of cancer-associated genomic regions and may interrogate hundreds of genes associated with cancer biology, DNA repair, copy-number instability, structural rearrangements, methylation states, and transcriptional dysregulation. In some embodiments, the targeted sequencing platform profiles approximately 600-800 genes and is capable of detecting single nucleotide variants (SNVs), insertions and deletions (indels), copy-number alterations (CNAs), structural rearrangements, gene fusions, epigenetic biomarkers, and transcriptomic biomarkers.

[0021] The disclosure further provides ensemble-based computational systems configured to determine probabilistic HRD scores using multiple orthogonal classes of genomic and epigenomic features. In certain embodiments, the systems generate HRD predictions using combinations of SNV mutational signatures, copy -number variation metrics, genomic scarring metrics, methylation features, transcriptomic features, tumor fraction estimates, ploidy estimates, and quality control metrics. In particular embodiments, the disclosed systems utilize machine learning classifiers including logistic regression models, ensemble learning approaches, neural networks, support vector machines, Bayesian classifiers, random forest classifiers, and related computational approaches.

[0022] In various embodiments, the disclosed methods utilize mutational signature projection approaches that decompose observed mutational patterns into biologically relevant mutational signatures. In certain embodiments, targeted sequencing-derived SNVs are converted into normalized mutational count matrices corresponding to trinucleotide mutational contexts and are subsequently projected onto reference mutational signatures including COSMIC mutational signatures. The disclosed systems may identify signatures associated with homologous recombination deficiency, mismatch repair deficiency, clocklike mutational processes, and additional biological processes relevant to tumor evolution and therapeutic responsiveness.

[0023] In some embodiments, the systems incorporate one or more HRD-associated mutational signatures including SBS3 and SBS39. In certain embodiments, mismatch repair deficiency-associated signatures including SBS6, SBS15, and SBS26 are negatively associated with HRD and are incorporated into classifier architectures to improve specificity. In some implementations, projection scores are determined using matrix decomposition methods including non-negative least squares (NNLS), non-negative matrix factorizationAttorney Docket No. GH0286WO(NMF), singular value decomposition, principal component analysis, or combinations thereof.

[0024] The present disclosure additionally provides systems for improving the specificity and robustness of HRD determination in low-input sequencing environments, including liquid biopsy samples containing circulating tumor DNA (ctDNA). In various embodiments, the methods incorporate copy-number quality control metrics configured to identify samples exhibiting unstable ploidy estimation, poor CNV model fit, or noisy segmentation profiles. In certain embodiments, samples failing quality thresholds are classified as not evaluable. In some implementations, the systems compute quality metrics including mean copy-number distance and mean minor allele frequency (MAF) distance to assess concordance between observed and predicted genomic states.

[0025] In certain embodiments, the disclosed systems further improve HRD determination by adjusting genomic scarring features according to estimated tumor ploidy. For example, large-scale state transition (LST) values may be normalized using ploidy-specific correction factors or polyploid normalization methods to reduce false-positive HRD calls associated with highly rearranged polyploid tumors.

[0026] In various embodiments, the disclosed ensemble models selectively apply different classifier architectures depending on the amount of genomic information available from a sample. In some embodiments, samples containing at least a threshold number of somatic SNVs are analyzed using a combined CNV+SNV model, while samples below the threshold utilize a CNV-only model. Such approaches improve stability and enable reliable operation across diverse tumor fractions and sequencing conditions.

[0027] The disclosed technologies additionally provide clinically actionable methods for selecting or administering therapies based on HRD status. In various embodiments, the methods are used to identify patients likely to respond to DNA damage response-targeted therapies including PARP inhibitors, platinum agents, ATR inhibitors, DNA-PK inhibitors, and related therapies. In certain embodiments, the systems are configured to predict progression-free survival, radiographic progression-free survival, overall survival, treatment responsiveness, and likelihood of therapeutic benefit.

[0028] The disclosure further provides methods for identifying HRD-positive tumors lacking pathogenic BRCA1 or BRCA2 mutations, thereby expanding patient populations eligible for targeted therapeutic intervention beyond conventional mutation-only testing approaches. In some embodiments, the disclosed technologies enable longitudinal monitoring of HRD status using both tissue-derived and liquid biopsy-derived nucleic acids.Attorney Docket No. GH0286WO

[0029] In various embodiments, the disclosed methods utilize targeted next-generation sequencing data derived from tissue samples, liquid biopsy samples, or combinations thereof. The targeted sequencing assays are configured to profile a plurality of cancer-associated genomic regions and may interrogate hundreds of genes associated with cancer biology, DNA repair, copy-number instability, structural rearrangements, methylation states, and transcriptional dysregulation. In some embodiments, the targeted sequencing platform profiles approximately 600-800 genes and is capable of detecting single nucleotide variants (SNVs), insertions and deletions (indels), copy-number alterations (CNAs), structural rearrangements, gene fusions, epigenetic biomarkers, and transcriptomic biomarkers.

[0030] The disclosure further provides ensemble-based computational systems configured to determine probabilistic HRD scores using multiple orthogonal classes of genomic and epigenomic features. In certain embodiments, the systems generate HRD predictions using combinations of SNV mutational signatures, copy -number variation metrics, genomic scarring metrics, methylation features, transcriptomic features, tumor fraction estimates, ploidy estimates, and quality control metrics. In particular embodiments, the disclosed systems utilize machine learning classifiers including logistic regression models, ensemble learning approaches, neural networks, support vector machines, Bayesian classifiers, random forest classifiers, and related computational approaches.

[0031] In various embodiments, the disclosed methods utilize mutational signature projection approaches that decompose observed mutational patterns into biologically relevant mutational signatures. In certain embodiments, targeted sequencing-derived SNVs are converted into normalized mutational count matrices corresponding to trinucleotide mutational contexts and are subsequently projected onto reference mutational signatures including COSMIC mutational signatures. The disclosed systems may identify signatures associated with homologous recombination deficiency, mismatch repair deficiency, clocklike mutational processes, and additional biological processes relevant to tumor evolution and therapeutic responsiveness.

[0032] In some embodiments, the systems incorporate one or more HRD-associated mutational signatures including SBS3 and SBS39. In certain embodiments, mismatch repair deficiency-associated signatures including SBS6, SBS15, and SBS26 are negatively associated with HRD and are incorporated into classifier architectures to improve specificity. In some implementations, projection scores are determined using matrix decomposition methods including non-negative least squares (NNLS), non-negative matrix factorizationAttorney Docket No. GH0286WO(NMF), singular value decomposition, principal component analysis, or combinations thereof.

[0033] The present disclosure additionally provides systems for improving the specificity and robustness of HRD determination in low-input sequencing environments, including liquid biopsy samples containing circulating tumor DNA (ctDNA). In various embodiments, the methods incorporate copy-number quality control metrics configured to identify samples exhibiting unstable ploidy estimation, poor CNV model fit, or noisy segmentation profiles. In certain embodiments, samples failing quality thresholds are classified as not evaluable. In some implementations, the systems compute quality metrics including mean copy-number distance and mean minor allele frequency (MAF) distance to assess concordance between observed and predicted genomic states.

[0034] In certain embodiments, the disclosed systems further improve HRD determination by adjusting genomic scarring features according to estimated tumor ploidy. For example, large-scale state transition (LST) values may be normalized using ploidy-specific correction factors or polyploid normalization methods to reduce false-positive HRD calls associated with highly rearranged polyploid tumors.

[0035] In various embodiments, the disclosed ensemble models selectively apply different classifier architectures depending on the amount of genomic information available from a sample. In some embodiments, samples containing at least a threshold number of somatic SNVs are analyzed using a combined CNV+SNV model, while samples below the threshold utilize a CNV-only model. Such approaches improve stability and enable reliable operation across diverse tumor fractions and sequencing conditions.

[0036] The disclosed technologies additionally provide clinically actionable methods for selecting or administering therapies based on HRD status. In various embodiments, the methods are used to identify patients likely to respond to DNA damage response-targeted therapies including PARP inhibitors, platinum agents, ATR inhibitors, DNA-PK inhibitors, and related therapies. In certain embodiments, the systems are configured to predict progression-free survival, radiographic progression-free survival, overall survival, treatment responsiveness, and likelihood of therapeutic benefit.

[0037] The disclosure further provides methods for identifying HRD-positive tumors lacking pathogenic BRCA1 or BRCA2 mutations, thereby expanding patient populations eligible for targeted therapeutic intervention beyond conventional mutation-only testing approaches. In some embodiments, the disclosed technologies enable longitudinal monitoring of HRD status using both tissue-derived and liquid biopsy-derived nucleic acids.Attorney Docket No. GH0286WO

[0038] In various embodiments, the present disclosure provides systems, methods, computer-implemented methods, software, sequencing assays, classifiers, and kits for determining homologous recombination deficiency (HRD) and / or genomic instability status (GIS) in a biological sample.

[0039] In certain embodiments, a targeted NGS assay profiles hundreds of cancer-associated genes for single nucleotide variants (SNVs), insertions and deletions (indels), copy-number alterations (CNAs), structural variants and gene fusions, methylation biomarkers, transcriptomic biomarkers, and mutational signatures. In some embodiments, the assay comprises a targeted sequencing panel comprising at least 200 genes, at least 300 genes, at least 500 genes, or approximately 600-800 genes.

[0040] In certain embodiments, HRD determination comprises generating a probabilistic score using an ensemble classifier incorporating SNV mutational signatures, genomic scarring metrics, copy-number features, epigenetic features, transcriptomic features, tumor fraction estimates, ploidy estimates, and quality control metrics. In some embodiments, the ensemble classifier selectively applies a CNV-only model or a combined CNV+SNV model based on whether a sample contains a threshold number of somatic SNVs.

[0041] In certain embodiments, the SNV component comprises projection of observed mutational contexts onto reference mutational signatures, including COSMIC mutational signatures. In some embodiments, the mutational signatures comprise one or more of SBS3, SBS39, SBS6, SBS15, SBS26, SBS5, and additional biologically relevant signatures.

[0042] In certain embodiments, the disclosure provides methods for improving HRD prediction by adjusting LST scores based on tumor ploidy, filtering samples using copynumber modeling quality metrics, excluding samples exhibiting poor CNV fit, incorporating mismatch repair deficiency signatures negatively associated with HRD, using normalized mutational count matrices, and using probabilistic logistic regression ensemble models.

[0043] In some embodiments, the systems generate a binary HRD call selected from HRD detected, HRD not detected, or HRD not evaluable. In various embodiments, the disclosure provides methods for selecting or administering therapy based on HRD status. In certain embodiments, the therapy comprises a PARP inhibitor, platinum chemotherapy, an ATR inhibitor, a DNA-PK inhibitor, an immune checkpoint inhibitor, or a combination thereof.

[0044] In certain embodiments, the disclosure provides methods for predicting progression-free survival (PFS), radiographic progression-free survival (rPFS), overall survival (OS), time to PSA progression, likelihood of treatment response, and therapeutic sensitivity. In some embodiments, the methods identify HRD-positive patients lacking pathogenic BRCA1 / 2Attorney Docket No. GH0286WOmutations. In some embodiments, the disclosure provides methods for dual tissue and liquid biopsy HRD monitoring.BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1. HRD score ensemble model workflow. Depicted herein is a workflows that illustrates an exemplary ensemble workflow for determining genomic instability status using combined SNV and CNV biomarkers. SBS Breast: Probability that SNV signature of sample is derived from that of breast HRD+ samples. SBS Prostate: Probability that SNV signature of sample is derived from that of prostate HRD+ samples. LOH score: number of segments shorter than 90% of the chromosome length with the loss of one allele. LST score: number of chromosomal breaks of adjacent regions at least 10 MB in size. TAI score: number of segments with allelic imbalance that extend to the subtelomere (within 5 MB of telomere ends), but do not cross the centromere.

[0046] Figure 2. Tissue Orthogonal HRD Score Concordance illustrates selection of biologically relevant mutational signatures for inclusion in an HRD classifier. Tissue Orthogonal HRD Score Concordance. Small cohort (n=l 5) of mCRPC patients with HRR mutations treated with PARPi rucaparib (TRITON2 trial) analyzed prior to opening up full cohort (n=145) for clinical outcomes analysis. Orthogonal HRD score from other method (%gLOH) analyzed to assess potential strong outliers, correlation due to method differences not expected. Strong correlation between GH equivalent of %gLOH and tissue-based %gLOH score from other method observed. One outlier highlighted sample with poor QC for copy number calling which resulted in many FP LoH events.

[0047] Figure 3. Association with clinical outcomes and HRR mutation status. All patients enrolled with BRCA1 / 2 or PALB2 mutations and TF > LoD detected as HRD All BRCA / PALB2 mutations germline or homozygous deletions. Two potential false positives (non-BRCA / PALB2-mut) in patients with poor outcomes identified that weakened association with radiographic progression-free survival (rPFS). Pt 1 with CDK12 mutation with poor CNV QC resulted in inflated LOH score. Pt 2 with FANCA mutation, tetrapioid sample with inflated LST score.

[0048] Figure 4. HRD Score Training, Validation, Thresholding illustrates expansion of HRD detection beyond BRCAl / 2-mutated patient populations. HRR-wildtype label for Training / Testing, No deletions (LOH / homdel) or promoter methylation in 14 HRR genes,

[0049] No germline pathogenic mutations in 14 HRR genes, No non-synonymous / frameshift somatic mutations in 14 HRR genes: BRCA1, BRCA2, ATM, BARD1, BRTPl, CDK12,Attorney Docket No. GH0286WOCHEK1, CHEK2, FANCL, PALB2, RAD51B, RAD51C, RAD51D, RAD54L. Strict HRD-Positive Label for Testing: Biallelic loss in BRCA1 or BRCA2: germline deleterious + LOH* or somatic deleterious + LOH*, somatic** deleterious + germline deleterious, 2 somatic** deleterious, Homozygous deletions, BRCA1 PM + LOH, *LOH required to have tumor CN=1, **Somatic deleterious mutations must be > 0.1% VAF or 10% of mTF if higher (considered clonal), Lenient HRD-Positive Label for Training (to allow for larger training set): HomDel Promoter methylation + LoH (in just RAD51C or BRCA1), Clonal deleterious (annotated by HRD module) SNV / indel / fusion OR biallelic (using BIP mutant allele status classification), Genes: BRCA1, BRCA2, RAD51D, RAD51C, PALB2.

[0050] Figure 5. COSMIC SBS signatures and relevance to HRD. Single base substitutions (SBS) signatures identified using 96 different contexts, considering the mutated base (six substitution subtypes: C>A, C>G, C>T, T>A, T>C, and T>G) and the bases immediately 5’ and 3’. Signature 3 (SBS3) is linked to defective homologous recombination DNA repair and is strongly associated with both germline and somatic biallelic inactivation of BRCA1 and BRCA2 genes. In breast cancer Signature 3 outperforms LOH and LST scores in identifying samples with events in HR-pathway genes: BRCA1, BRCA2, RAD51C, and PALB2 (WGS, tissue). In pancreatic, breast and ovarian cancer, responders to platinum therapy and PARPi usually exhibit SB S3 mutations. Previously, a de novo mutational signature was derived from HRD-positive Infinity samples. To improve SNV score stability and potential overfitting, a simplified approach using existing COSMIC signatures was implemented Signatures that have statistically significant difference between HRD+ and HRD- training samples were selected for inclusion in the projection to generate a new SNV signature score P-values computed using t-test with BH correction Selected 6 signatures with known etiology and unrelated to chemotherapy or carcinogen exposure, including SB S3 (HRD) and SBS6, SBS15, and SBS26 (DNA mismatch repair). SBS3 scores were slightly increased in HRD-positives, while dMMR signature scores were decreased, consistent with the mutual exclusivity of HRD and MMR deficiency.

[0051] Figure 6. New SNV score has much higher stability. With > 20 SNVs determined as optimal cut-off based on ensemble model performance illustrates progression-free survival associations with GIS classification.

[0052] Figure 7. LST ploidy-based adjustment using multiple normalization factors.Normalizing LST scores from polyploid samples (based on medians) improves performance in breast prostate ovarian and pancreatic samples. Vast majority of samples are diploid or tetrapioid, limited samples to capture LST score distribution for higher ploidy. May partiallyAttorney Docket No. GH0286WOexplain why median LST scores are non-monotonic with increased ploidy. Comparison revealed some instability in ploidy estimations (<3% of S3 / LY pairs with discordances)

[0053] Figure 8. Normalization of LST scores with a single polyploid factor further improves performance. Given ploidy instability and uneven distribution of data across ploidies, explored using a single normalization factor for polyploid samples. A single polyploid factor resulted in marginally better performance in breast ovarian prostate and pancreatic samples (0.9 AUC with multiple polyploidy LST normalization factors).

[0054] Figure 9. Using difference between observed and predicted CN / MAFs as CNV QC filter improves performance.

[0055] Figure 10. Revised ensemble model workflow. Features SBS*: SNV signature score derived from sample mutational count matrix projection onto biologically relevant COSMIC signatures. LOH score: number of segments shorter than 90% of the chromosome length with the loss of one allele. LST score: number of chromosomal breaks of adjacent regions at least 10 MB in size, adjusted for polyploidy. TAI score: number of segments with allelic imbalance that extend to the sub telomere (within 5 MB of telomere ends), but do not cross the centromere.

[0056] Figure 11. GIS expands HRD detection to patients without pathogenic BRCA1 / 2 mutations. To estimate the overall prevalence of GIS detection and the HRR genotypes of the detected cases we explored a cohort of clinical Guardant360 Tissue samples. We detected HRD through GIS in about one third of breast patients, 45% of ovarian patients, and 14% of pancreatic patients. Notably, among the GIS-detected cases only about half had BRCA1 / 2 pathogenic mutations across all 3 cancer types, and we identified an expanded HRD patient population for therapy selection shown in the medium blue color. These patients displayed genomic instability in the absence of BRCA1 / 2 pathogenic mutations. A small percentage did harbor pathogenic mutations in other HRR genes, but the remaining patients had HRR alterations of unclear significance.DETAILED DESCRIPTION

[0057] As described, conventional approaches for identifying HRD have largely focused on detection of deleterious germline or somatic mutations in HRR genes such as BRCA1, BRCA2, PALB2, RAD51C, RAD51D, ATM, BRIP1, BARD1, CHEK1, CHEK2, and related genes. However, mutation-only approaches incompletely capture the full spectrum of tumors exhibiting functional HRD. A substantial subset of tumors lacking canonical BRCA1 or BRCA2 mutations nevertheless exhibit genomic scarring patterns and mutational signaturesAttorney Docket No. GH0286WOcharacteristic of impaired homologous recombination repair. In addition, tumors may acquire genomic instability through transient or historical HRR dysfunction and subsequently retain genomic scarring signatures even after partial restoration of repair capacity.

[0058] As a consequence, increasing attention has been directed toward genomic instability biomarkers capable of identifying functional HRD independently of mutation status alone. Such biomarkers include loss of heterozygosity (LOH), large-scale state transitions (LST), telomeric allelic imbalance (TAI), mutational signatures including COSMIC SBS3, chromosomal rearrangement patterns, methylation biomarkers, and transcriptomic features associated with defective DNA repair. Existing tissue-based HRD assays have demonstrated clinical utility in certain settings; however, many currently available approaches rely on whole-genome sequencing, large genomic arrays, or analytical frameworks that are not readily transferable to targeted sequencing assays or liquid biopsy workflows.

[0059] Detection of HRD in circulating tumor DNA and other low-input biological samples presents substantial technical challenges. Cell-free DNA samples often contain limited quantities of tumor-derived nucleic acids and correspondingly small numbers of detectable somatic variants. Copy-number estimation in liquid biopsy samples may be affected by low tumor fraction, ploidy instability, sequencing noise, and segmentation artifacts. Moreover, mutational signature methods originally developed using whole-genome sequencing data frequently exhibit reduced stability or reduced sensitivity when applied to targeted sequencing assays, particularly in samples containing low mutation burdens.

[0060] Additional challenges arise from tumor heterogeneity and differences between tumor types. For example, prostate cancers may exhibit distinct genomic scarring profiles compared to ovarian or breast cancers, while highly polyploid tumors may artificially inflate certain genomic instability metrics and generate false-positive HRD classifications. Similarly, mismatch repair deficiency and other mutational processes may confound mutational signature analysis unless appropriately modeled. Additional information is found in PCT App. No. PCT / US2024 / 034033 and PCT Pub. No. WO2024 / 259251A1, which are fully incorporated by reference herein in their entirety.

[0061] In various embodiments, a biological sample is obtained from a subject. The biological sample may comprise any sample containing cell-free DNA, including but not limited to plasma, serum, whole blood, or other bodily fluids. In some embodiments, plasma is preferred due to its established utility in cfDNA-based assays. The cfDNA is then isolated or otherwise made available for downstream methylation analysis using methods known in the art.Attorney Docket No. GH0286WO

[0062] The methods disclosed herein involve the measurement of DNA methylation at a plurality of genomic regions that are functionally defined as lung tissue-specific differentially methylated regions (DMRs). These regions are not limited to specific genomic coordinates but are instead defined by their functional properties. In particular, such regions exhibit differential methylation patterns in lung tissue relative to non-lung tissues, including blood-derived cells and other organ systems. In certain embodiments, the regions are characterized by relatively high methylation levels in lung tissue and comparatively lower methylation levels in non-lung tissues, or vice versa.

[0063] In an embodiment, the population of nucleic acids is one obtained from a serum, plasma or blood sample from a subject suspected of having neoplasia, a tumor, or cancer or previously diagnosed with neoplasia, a tumor, or cancer. The population of nucleic acids includes nucleic acids having varying levels of methylation. Methylation can occur from any one or more post-replication or transcriptional modifications. Post-replication modifications include modifications of the nucleotide cytosine, particularly at the 5-position of the nucleobase, e.g., 5-methylcytosine, 5-hydroxymethylcytosine, 5-formylcytosine and 5-carboxylcytosine. The affinity agents can be antibodies with the desired specificity, natural binding partners or variants thereof (Bock et al., Nat Biotech 28: 1106-1114 (2010); Song et al., Nat Biotech 29: 68-72 (2011)), or artificial peptides selected e.g., by phage display to have specificity to a given target.

[0064] Examples of capture moieties contemplated herein include methyl binding domain (MBDs) and methyl binding proteins (MBPs) as described herein, including proteins such as MeCP2 and antibodies preferentially binding to 5-methylcytosine. Likewise, partitioning of different forms of nucleic acids can be performed using histone binding proteins which can separate nucleic acids bound to histones from free or unbound nucleic acids. Examples of histone binding proteins that can be used in the methods disclosed herein include RBBP4, RbAp48 and SANT domain peptides. Although for some affinity agents and modifications, binding to the agent may occur in an essentially all or none manner depending on whether a nucleic acid bears a modification, the separation may be one of degree. In such instances, nucleic acids overrepresented in a modification bind to the agent at a greater extent that nucleic acids underrepresented in the modification. Alternatively, nucleic acids having modifications may bind in an all or nothing manner. But then, various levels of modifications may be sequentially eluted from the binding agent.

[0065] For example, in some embodiments, partitioning can be binary or based on degree / level of modifications. For example, all methylated fragments can be partitioned fromAttorney Docket No. GH0286WOunmethylated fragments using methyl-binding domain proteins (e.g., MethylMiner Methylated DNA Enrichment Kit (ThermoFisher Scientific)). Subsequently, additional partitioning may involve eluting fragments having different levels of methylation by adjusting the salt concentration in a solution with the methyl-binding domain and bound fragments. In some embodiments, partitioning of the nucleic acids is performed by contacting the nucleic acids with a methylation binding domain (“MBD”) of a methylation binding protein (“MBP”). MBD binds to 5-methylcytosine (5mC). MBD is coupled to paramagnetic beads, such as Dynabeads® M-280 Streptavidin via a biotin linker. Partitioning into fractions with different extents of methylation can be performed by eluting fractions by increasing the NaCl concentration.

[0066] An exemplary method for molecular tag identification of MBD-bead partitioned libraries through NGS is as follows:

[0067] Physical partitioning of an extracted DNA sample (e.g., extracted blood plasma DNA from a human sample) using a methyl-binding domain protein-bead purification kit, saving all elutions from process for downstream processing.

[0068] Parallel application of differential molecular tags and NGS-enabling adapter sequences to each partition. For example, the hypermethylated, residual methylation ('wash'), and hypomethylated partitions are ligated with NGS-adapters with molecular tags.

[0069] Re-combining all molecular tagged partitions, and subsequent amplification using adapter-specific DNA primer sequences.

[0070] Enrichment / hybridization of re-combined and amplified total library, targeting genomic regions of interest (e.g., cancer-specific genetic variants and differentially methylated regions).

[0071] Re-amplification of the enriched total DNA library, appending a sample tag.Different samples are pooled and assayed in multiplex on an NGS instrument.

[0072] Bioinformatics analysis of NGS data, with the molecular tags being used to identify unique molecules, as well deconvolution of the sample into molecules that were differentially MBD-partitioned. This analysis can yield information on relative 5-methylcytosine for genomic regions, concurrent with standard genetic sequencing / variant detection.

[0073] Examples of MBPs contemplated herein include, but are not limited to:(a) MeCP2 is a protein preferentially binding to 5-methyl-cytosine over unmodified cytosine.(b) RPL26, PRP8 and the DNA mismatch repair protein MHS6 preferentially bind to 5- hydroxymethyl-cytosine over unmodified cytosine.Attorney Docket No. GH0286WO(c) F0XK1, FOXK2, FOXP1, FOXP4 and FOXI3 preferably bind to 5-formyl- cytosine over unmodified cytosine (lurlaro et al., Genome Biol. 14: R119 (2013)). (d) Antibodies specific to one or more methylated nucleotide bases.

[0074] In some embodiments, the lung-specific DMRs are identified through a multi-step selection process. This process may include analyzing methylation data across multiple tissue types, including lung tissue and non-lung tissues such as blood cells, to identify candidate regions exhibiting tissue-specific methylation patterns. Statistical methods, including differential methylation analysis using false discovery rate thresholds and minimum methylation differences, may be employed to identify regions that are significantly differentially methylated. Additional filtering steps may be applied to remove regions with high background methylation in circulating DNA or regions associated with unrelated cancer types, for example by using reference datasets such as TCGA. In certain embodiments, regions are selected that contain a minimum number of CpG sites and satisfy predefined methylation thresholds in target versus non-target tissues.

[0075] Using such approaches, a set of lung-specific DMRs may be identified and used as features for downstream analysis. In some embodiments, at least 10 such regions are used. In other embodiments, the number of regions may range from approximately 15 to 30. In certain specific embodiments, a subset of approximately 19 lung-specific DMRs is used as a feature set for detecting lung injury. These regions may include hypermethylated regions associated with lung tissue, which have been observed to exhibit increased methylation signals in subjects with lung injury.

[0076] Once methylation levels at the selected regions are measured, a methylation profile is generated for the sample. Measurement of methylation may be performed using a variety of platforms. In some embodiments, bisulfite sequencing is employed, including whole-genome bisulfite sequencing or targeted bisulfite sequencing approaches. In other embodiments, targeted sequencing panels are used to selectively interrogate the regions of interest. In still other embodiments, methylation-sensitive restriction enzyme (MSRE)-based assays are employed, in which enzymatic digestion patterns are used to infer methylation states at specific genomic loci. The invention is not limited to any particular methylation detection platform, and any suitable method capable of distinguishing methylated from unmethylated DNA may be used.

[0077] The resulting methylation profile may then be analyzed to determine whether the subject exhibits lung injury. In some embodiments, the analysis involves unsupervised learning approaches. For example, clustering algorithms such as k-means clustering,Attorney Docket No. GH0286WOGaussian mixture models (GMM), principal component analysis (PCA), or uniform manifold approximation and projection (UMAP) may be used to group samples based on similarities in methylation patterns. In certain embodiments, samples segregate into distinct clusters corresponding to high and low methylation states across the lung-specific DMRs. Samples exhibiting higher methylation signals at these regions have been observed to correspond to increased levels of lung-derived cfDNA and are associated with lung injury.

[0078] In other embodiments, supervised machine learning models are used to classify samples. Such models may include, but are not limited to, logistic regression models, support vector machines, random forests, or neural networks. In some embodiments, a logistic regression model is trained using labeled samples to distinguish between subjects with and without lung injury. The model may be trained using cross-validation techniques, such as 10-fold cross-validation, to ensure robustness and generalizability. Performance metrics including sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) may be used to evaluate the model. In certain embodiments, models trained using the disclosed lung-specific DMR features achieve high performance, including AUC values of approximately 0.96 and sensitivity and specificity values exceeding 90%.

[0079] In addition to methylation features, other parameters may be incorporated into the analysis. For example, total cfDNA concentration or yield may be used as an additional feature, as increased cfDNA levels have been observed in subjects with lung injury. In certain embodiments, subjects classified as having lung injury exhibit both elevated methylation signals at lung-specific DMRs and increased cfDNA levels relative to baseline or control subjects.Samples

[0080] Isolation and extraction of cell free polynucleotides may be performed through collection of samples using a variety of techniques. A sample can be any biological sample isolated from a subject. Samples can include body tissues, whole blood, platelets, serum, plasma, stool, red blood cells, white blood cells or leucocytes, endothelial cells, tissue biopsies (e.g., biopsies from known or suspected solid tumors), cerebrospinal fluid, synovial fluid, lymphatic fluid, ascites fluid, interstitial or extracellular fluid (e.g., fluid from intercellular spaces), gingival fluid, crevicular fluid, bone marrow, pleural effusions,Attorney Docket No. GH0286WOcerebrospinal fluid, saliva, mucous, sputum, semen, sweat, urine. Samples are preferably body fluids, particularly blood and fractions thereof, and urine. Such samples include nucleic acids shed from tumors. The nucleic acids can include DNA and RNA and can be in double and single-stranded forms. A sample can be in the form originally isolated from a subject or can have been subjected to further processing to remove or add components, such as cells, enrich for one component relative to another, or convert one form of nucleic acid to another, such as RNA to DNA or single-stranded nucleic acids to double-stranded. Thus, for example, a body fluid sample for analysis is plasma or serum containing cell-free nucleic acids, e.g., cell-free DNA (cfDNA).

[0081] In some implementations, the sample volume of body fluid taken from a subject depends on the desired read depth for sequenced regions. Example volumes are about 0.4-40 ml, about 5-20 ml, about 10-20 ml. For example, the volume can be about 0.5 ml, about 1 ml, about 5 ml, about 10 ml, about 20 ml, about 30 ml, about 40 ml, or more milliliters. A volume of sampled blood can be between about 5 ml to about 20 ml.

[0082] The sample can comprise various amounts of nucleic acid. The amount of nucleic acid in a given sample can be equated with multiple genome equivalents. For example, a sample of about 30 ng DNA can contain about 10,000 (104) haploid human genome equivalents and, in the case of cfDNA, about 200 billion (2x1011) individual polynucleotide molecules.Similarly, a sample of about 100 ng of DNA can contain about 30,000 haploid human genome equivalents and, in the case of cfDNA, about 600 billion individual molecules.

[0083] In some implementations, a sample comprises nucleic acids from different sources, e.g., from cells and from cell-free sources (e.g., blood samples, etc.). Typically, a sample includes nucleic acids carrying mutations. For example, a sample optionally comprises DNA carrying germline mutations and / or somatic mutations. Typically, a sample comprises DNA carrying cancer-associated mutations (e.g., cancer-associated somatic mutations). In some implementations of the present disclosure, cell free nucleic acids in a subject may derive from a tumor. For example, cell-free DNA isolated from a subject can comprise ctDNA.

[0084] Example amounts of cell-free nucleic acids in a sample before amplification typically range from about 1 femtogram (fg) to about 1 microgram (pg), e.g., about 1 picogram (pg) to about 200 nanogram (ng), about 1 ng to about 100 ng, about 10 ng to about 1000 ng. In some implementations, a sample includes up to about 600 ng, up to about 500 ng, up to about 400 ng, up to about 300 ng, up to about 200 ng, up to about 100 ng, up to about 50 ng, or up to about 20 ng of cell-free nucleic acid molecules. Optionally, the amount is at least about 1 fg, at least about 10 fg, at least about 100 fg, at least about 1 pg, at least about 10 pg, at leastAttorney Docket No. GH0286WOabout 100 pg, at least about 1 ng, at least about 10 ng, at least about 100 ng, at least about 150 ng, or at least about 200 ng of cell-free nucleic acid molecules. In certain implementations, the amount is up to about 1 fg, about 10 fg, about 100 fg, about 1 pg, about 10 pg, about 100 pg, about 1 ng, about 10 ng, about 100 ng, about 150 ng, or about 200 ng of cell-free nucleic acid molecules. In some implementations, methods include obtaining between about 1 fg to about 200 ng cell-free nucleic acid molecules from samples.

[0085] Cell-free nucleic acids typically have a size distribution of between about 100 nucleotides in length and about 500 nucleotides in length, with molecules of about 110 nucleotides in length to about 230 nucleotides in length representing about 90% of molecules in the sample, with a mode of about 168 nucleotides length and a second minor peak in a range between about 240 to about 440 nucleotides in length. In certain implementations, cell-free nucleic acids are from about 160 to about 180 nucleotides in length, or from about 320 to about 360 nucleotides in length, or from about 440 to about 480 nucleotides in length.

[0086] In some implementations, cell-free nucleic acids are isolated from bodily fluids through a partitioning step in which cell-free nucleic acids, as found in solution, are separated from intact cells and other non-soluble components of the bodily fluid. In some of these implementations, partitioning includes techniques such as centrifugation or filtration.Alternatively, cells in bodily fluids are lysed, and cell-free and cellular nucleic acids processed together. Generally, after addition of buffers and wash steps, cell-free nucleic acids are precipitated with, for example, an alcohol. In certain implementations, additional clean up steps are used, such as silica-based columns to remove contaminants or salts. Non-specific bulk carrier nucleic acids, for example, are optionally added throughout the reaction to optimize certain aspects of the example procedure, such as yield. After such processing, samples typically include various forms of nucleic acids including double-stranded DNA, single-stranded DNA and / or single-stranded RNA. Optionally, single stranded DNA and / or single stranded RNA are converted to double stranded forms so that they are included in subsequent processing and analysis steps.

[0087] The disclosed methods may be performed using tissue-derived nucleic acids, cell-free nucleic acids, circulating tumor DNA, circulating tumor RNA, or combinations thereof. In certain embodiments, the biological sample comprises a plasma sample obtained from a subject having cancer. In some embodiments, the biological sample comprises formalin-fixed paraffin-embedded tissue. Additional sample types may include serum, whole blood, urine, cerebrospinal fluid, pleural fluid, ascites, saliva, and related clinical specimens.Attorney Docket No. GH0286WO

[0088] In certain embodiments, nucleic acids are isolated and enriched using molecular barcoding methods, methyl-binding domain enrichment approaches, targeted capture probes, hybridization enrichment methods, multiplex PCR amplification, adapter ligation, and sequencing library preparation techniques known in the art.

[0089] The disclosed methods are particularly advantageous for targeted sequencing workflows utilizing low-input nucleic acid quantities, including cell-free DNA samples exhibiting low tumor fractions. In various embodiments, sequencing libraries are generated using molecular identifiers to improve sensitivity and reduce sequencing artifacts. Additional details regarding cfDNA partitioning and related analysis of epigenetic modifications that are optionally adapted for use in performing the methods disclosed herein are described in, for example, WO 2018 / 119452, filed December 22, 2017, which is incorporated by reference.Sequencing Panel

[0090] To improve the likelihood of detecting genomic regions of interest and optionally, tumor indicating mutations, the sections of DNA sequenced may comprise a panel of genes or genomic sections that comprise known genomic regions. Selection of a limited section for sequencing (e.g., a limited panel) can reduce the total sequencing needed (e.g., a total amount of nucleotides sequenced). A sequencing panel can target a plurality of different genes or regions, for example, to detect a single cancer, a set of cancers, or all cancers. Alternatively, DNA may be sequenced by whole genome sequencing (WGS) or other unbiased sequencing method without the use of a sequencing panel. Examples of suitable panel and targets for use in panels can be found in the epigenetic targets described in US provisional patent application 62 / 799,637, filed January 31, 2019, which is incorporated by reference in its entirety. Other examples include Mutational Signatures found in Catalogue of Somatic Mutations in Cancer (COSMIC) available at https: / / cancer.sanger.ac.uk / signatures / sbs / .

[0091] In some aspects, a panel that targets a plurality of different genes or genomic regions (e.g., transcriptional factor binding regions, distal regulatory elements (DREs), repetitive elements, intron-exon junctions, transcriptional start sites (TSSs), and / or the like) is selected such that a determined proportion of subjects having a cancer exhibits a genetic variant or tumor marker in one or more different genes in the panel. The panel may be selected to limit a region for sequencing to a fixed number of base pairs. The panel may be selected to sequence a desired amount of DNA. The panel may be further selected to achieve a desiredAttorney Docket No. GH0286WOsequence read depth. The panel may be selected to achieve a desired sequence read depth or sequence read coverage for an amount of sequenced base pairs. The panel may be selected to achieve a theoretical sensitivity, a theoretical specificity, and / or a theoretical accuracy for detecting one or more genetic variants in a sample.

[0092] Probes for detecting the panel of regions can include those for detecting genomic regions of interest (hotspot regions) as well as nucleosome-aware probes (e.g., KRAS codons 12 and 13) and may be designed to optimize capture based on analysis of cfDNA coverage and fragment size variation impacted by nucleosome binding patterns and GC sequence composition. Regions used herein can also include non-hotspot regions optimized based on nucleosome positions and GC models. The panel can comprise a plurality of subpanels, including subpanels for identifying tissue of origin (e.g., use of published literature to define 50-100 baits representing genes with most diverse transcription profile across tissues (not necessarily promoters)), whole genome scaffold (e.g., for identifying ultra-conservative genomic content and tiling sparsely across chromosomes with handful of probes for copy number base lining purposes), transcription start site (TSS) / CpG islands (e.g., for capturing differential methylated regions (e.g., Differentially Methylated Regions (DMRs)) in for example in promoters of tumor suppressor genes (e.g., SEPT9 / VIM in colorectal cancer)). In some implementations, markers for a tissue of origin are tissue-specific epigenetic markers.

[0093] In some implementations, the one or more regions in the panel comprise one or more loci from one or a plurality of genes for detecting residual cancer after surgery. This detection can be earlier than is possible for existing methods of cancer detection. In some implementations, the one or more genomic locations in the panel comprise one or more loci from one or a plurality of genes for detecting cancer in a high-risk patient population. For example, smokers have much higher rates of lung cancer than the general population.Moreover, smokers can develop other lung conditions that make cancer detection more difficult, such as the development of irregular nodules in the lungs. In some implementations, the methods described herein detect the response of patients to cancer therapy (particularly in high risk patients) earlier than is possible for existing methods of cancer detection.

[0094] A genomic location may be selected for inclusion in a sequencing panel based on a number of subjects with a cancer that have a tumor marker in that gene or region. A genomic location may be selected for inclusion in a sequencing panel based on prevalence of subjects with a cancer and a tumor marker present in that gene. Presence of a tumor marker in a region may be indicative of a subject having cancer.Attorney Docket No. GH0286WO

[0095] In some instances, the panel may be selected using information from one or more databases. The information regarding a cancer may be derived from cancer tumor biopsies or cfDNA assays. A database may comprise information describing a population of sequenced tumor samples. A database may comprise information about mRNA expression in tumor samples. A database may comprise information about regulatory elements or genomic regions in tumor samples. The information relating to the sequenced tumor samples may include the frequency of various genetic variants and describe the genes or regions in which the genetic variants occur. The genetic variants may be tumor markers. A non-limiting example of such a database is COSMIC. COSMIC is a catalogue of somatic mutations found in various cancers. For a particular cancer, COSMIC ranks genes based on frequency of mutation. A gene may be selected for inclusion in a panel by having a high frequency of mutation within a given gene. For instance, COSMIC indicates that 33% of a population of sequenced breast cancer samples have a mutation in TP53 and 22% of a population of sampled breast cancers have a mutation in KRAS. Other ranked genes, including APC, have mutations found only in about 4% of a population of sequenced breast cancer samples. TP53 and KRAS may be included in a sequencing panel based on having relatively high frequency among sampled breast cancers (compared to APC, for example, which occurs at a frequency of about 4%). COSMIC is provided as a non-limiting example, however, any database or set of information may be used that associates a cancer with tumor marker located in a gene or genetic region. In another example, as provided by COSMIC, of 1156 biliary tract cancer samples, 380 samples (33%) carried mutations in TP53. Several other genes, such as APC, have mutations in 4-8% of all samples. Thus, TP53 may be selected for inclusion in the panel based on a relatively high frequency in a population of biliary tract cancer samples.

[0096] A gene or genomic section may be selected for a panel where the frequency of a tumor marker is significantly greater in sampled tumor tissue or circulating tumor DNA than found in a given background population. A combination of genomic locations may be selected for inclusion of a panel such that at least a majority of subjects having a cancer may have a tumor marker or genomic region present in at least one of the genomic location or genes in the panel. The combination of genomic location may be selected based on data indicating that, for a particular cancer or set of cancers, a majority of subjects have one or more tumor markers in one or more of the selected regions. For example, to detect cancer 1, a panel comprising regions A, B, C, and / or D may be selected based on data indicating that 90% of subjects with cancer 1 have a tumor marker in regions A, B, C, and / or D of the panel. Alternately, tumor markers may be shown to occur independently in two or more regions inAttorney Docket No. GH0286WOsubjects having a cancer such that, combined, a tumor marker in the two or more regions is present in a majority of a population of subjects having a cancer. For example, to detect cancer 2, a panel comprising regions X, Y, and Z may be selected based on data indicating that 90% of subjects have a tumor marker in one or more regions, and in 30% of such subjects a tumor marker is detected only in region X, while tumor markers are detected only in regions Y and / or Z for the remainder of the subjects for whom a tumor marker was detected. Tumor markers present in one or more genomic locations previously shown to be associated with one or more cancers may be indicative of or predictive of a subject having cancer if a tumor marker is detected in one or more of those regions 50% or more of the time. Computational approaches such as models employing conditional probabilities of detecting cancer given a cancer frequency for a set of tumor markers within one or more regions may be used to predict which regions, alone or in combination, may be predictive of cancer. Other approaches for panel selection involve the use of databases describing information from studies employing comprehensive genomic profiling of tumors with large panels and / or whole genome sequencing (WGS, RNA-seq, Chip-seq, ATAC-seq, and others). Information gleaned from literature may also describe pathways commonly affected and mutated in certain cancers. Panel selection may be further informed by the use of ontologies describing genetic information.

[0097] Genes included in the panel for sequencing can include the fully transcribed region, the promoter region, enhancer regions, regulatory elements, and / or downstream sequence. To further increase the likelihood of detecting tumor indicating mutations only exons may be included in the panel. The panel can comprise all exons of a selected gene, or only one or more of the exons of a selected gene. The panel may comprise of exons from each of a plurality of different genes. The panel may comprise at least one exon from each of the plurality of different genes. In various embodiments, the disclosed systems utilize targeted next-generation sequencing assays configured to interrogate cancer-associated genomic loci. The targeted sequencing panels may include coding regions, non-coding regions, promoter regions, methylation-sensitive regions, transcriptomic targets, structural rearrangement hotspots, copy -number informative regions, or combinations thereof. In some embodiments, the sequencing panel interrogates at least 100 genes, at least 300 genes, at least 500 genes, or approximately 741 genes. In certain embodiments, the assay detects SNVs, indels, copynumber alterations, fusions, promoter methylation states, and transcriptomic biomarkers in a single integrated workflow. The targeted sequencing assays disclosed herein provide advantages over whole-genome sequencing approaches in liquid biopsy settings becauseAttorney Docket No. GH0286WOtargeted sequencing enables substantially greater sequencing depth across selected genomic regions, thereby improving sensitivity for low-frequency variants and low tumor fraction samples.

[0098] In some aspects, a panel of exons from each of a plurality of different genes is selected such that a determined proportion of subjects having a cancer exhibit a genetic variant in at least one exon in the panel of exons.

[0099] At least one full exon from each different gene in a panel of genes may be sequenced. The sequenced panel may comprise exons from a plurality of genes. The panel may comprise exons from 2 to 100 different genes, from 2 to 70 genes, from 2 to 50 genes, from 2 to 30 genes, from 2 to 15 genes, or from 2 to 10 genes.

[0100] A selected panel may comprise a varying number of exons. The panel may comprise from 2 to 3000 exons. The panel may comprise from 2 to 1000 exons. The panel may comprise from 2 to 500 exons. The panel may comprise from 2 to 100 exons. The panel may comprise from 2 to 50 exons. The panel may comprise no more than 300 exons. The panel may comprise no more than 200 exons. The panel may comprise no more than 100 exons. The panel may comprise no more than 50 exons. The panel may comprise no more than 40 exons. The panel may comprise no more than 30 exons. The panel may comprise no more than 25 exons. The panel may comprise no more than 20 exons. The panel may comprise no more than 15 exons. The panel may comprise no more than 10 exons. The panel may comprise no more than 9 exons. The panel may comprise no more than 8 exons. The panel may comprise no more than 7 exons.

[0101] The panel may comprise one or more exons from a plurality of different genes. The panel may comprise one or more exons from each of a proportion of the plurality of different genes. The panel may comprise at least two exons from each of at least 25%, 50%, 75% or 90% of the different genes. The panel may comprise at least three exons from each of at least 25%, 50%, 75% or 90% of the different genes. The panel may comprise at least four exons from each of at least 25%, 50%, 75% or 90% of the different genes.

[0102] The sizes of the sequencing panel may vary. A sequencing panel may be made larger or smaller (in terms of nucleotide size) depending on several factors including, for example, the total amount of nucleotides sequenced, or a number of unique molecules sequenced for a particular region in the panel. The sequencing panel can be sized 5 kb to 50 kb. The sequencing panel can be 10 kb to 30 kb in size. The sequencing panel can be 12 kb to 20 kb in size. The sequencing panel can be 12 kb to 60 kb in size. The sequencing panel can be at least lOkb, 12 kb, 15 kb, 20 kb, 25 kb, 30 kb, 35 kb, 40 kb, 45 kb, 50 kb, 60 kb, 70 kb,Attorney Docket No. GH0286WO80 kb, 90 kb, 100 kb, 110 kb, 120 kb, 130 kb, 140 kb, or 150 kb in size. The sequencing panel may be less than 100 kb, 90 kb, 80 kb, 70 kb, 60 kb, or 50 kb in size.

[0103] The panel selected for sequencing can comprise at least 1, 5, 10, 15, 20, 25, 30, 40, 50, 60, 80, or 100 genomic locations (e.g., that each include genomic regions of interest). In some cases, the genomic locations in the panel are selected that the size of the locations are relatively small. In some cases, the regions in the panel have a size of about 10 kb or less, about 8 kb or less, about 6 kb or less, about 5 kb or less, about 4 kb or less, about 3 kb or less, about 2.5 kb or less, about 2 kb or less, about 1.5 kb or less, or about 1 kb or less or less. In some cases, the genomic locations in the panel have a size from about 0.5 kb to about 10 kb, from about 0.5 kb to about 6 kb, from about 1 kb to about 11 kb, from about 1 kb to about 15 kb, from about 1 kb to about 20 kb, from about 0.1 kb to about 10 kb, or from about 0.2 kb to about 1 kb. For example, the regions in the panel can have a size from about 0.1 kb to about 5 kb.

[0104] The panel selected herein can allow for deep sequencing that is sufficient to detect low-frequency genetic variants (e.g., in cell-free nucleic acid molecules obtained from a sample). An amount of genetic variants in a sample may be referred to in terms of the mutant allele fraction for a given genetic variant. The mutant allele fraction may refer to the frequency at which mutant alleles occur in a given population of nucleic acids, such as a sample. Genetic variants at a low mutant allele fraction may have a relatively low frequency of presence in a sample. In some cases, the panel allows for detection of genetic variants at a mutant allele fraction of at least 0.0001%, 0.001%, 0.005%, 0.01%, 0.05%, 0.1%, or 0.5%. The panel can allow for detection of genetic variants at a mutant allele fraction of 0.001% or greater. The panel can allow for detection of genetic variants at a mutant allele fraction of 0.01% or greater. The panel can allow for detection of genetic variant present in a sample at a frequency of as low as 0.0001%, 0.001%, 0.005%, 0.01%, 0.025%, 0.05%, 0.075%, 0.1%, 0.25%, 0.5%, 0.75%, or 1.0%. The panel can allow for detection of tumor markers present in a sample at a frequency of at least 0.0001%, 0.001%, 0.005%, 0.01%, 0.025%, 0.05%, 0.075%, 0.1%, 0.25%, 0.5%, 0.75%, or 1.0%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 1.0%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.75%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.5%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.25%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.1%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.075%. The panelAttorney Docket No. GH0286WOcan allow for detection of tumor markers at a frequency in a sample as low as 0.05%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.025%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.01%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.005%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.001%. The panel can allow for detection of tumor markers at a frequency in a sample as low as 0.0001%. The panel can allow for detection of tumor markers in sequenced cfDNA at a frequency in a sample as low as 1.0% to 0.0001%. The panel can allow for detection of tumor markers in sequenced cfDNA at a frequency in a sample as low as 0.01% to 0.0001%.

[0105] A genetic variant can be exhibited in a percentage of a population of subjects who have a disease (e.g., cancer). In some cases, at least 1%, 2%, 3%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, or 99% of a population having the cancer exhibit one or more genetic variants in at least one of the regions in the panel. For example, at least 80% of a population having the cancer may exhibit one or more genetic variants in at least one of the genomic positions in the panel.

[0106] The panel can comprise one or more locations comprising genomic regions of interest from each of one or more genes. In some cases, the panel can comprise one or more locations comprising genomic regions of interest from each of at least 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, or 80 genes. In some cases, the panel can comprise one or more locations comprising genomic regions of interest from each of at most 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, or 80 genes. In some cases, the panel can comprise one or more locations comprising genomic regions of interest from each of from about 1 to about 80, from 1 to about 50, from about 3 to about 40, from 5 to about 30, from 10 to about 20 different genes.

[0107] The locations comprising genomic regions in the panel can be selected so that one or more epigenetically modified regions are detected. The one or more epigenetically modified regions can be acetylated, methylated, ubiquitylated, phosphorylated, sumoylated, ribosylated, and / or citrullinated. For example, the regions in the panel can be selected so that one or more methylated regions are detected.

[0108] The regions in the panel can be selected so that they comprise sequences differentially transcribed across one or more tissues. In some cases, the locations comprising genomic regions can comprise sequences transcribed in certain tissues at a higher levelAttorney Docket No. GH0286WOcompared to other tissues. For example, the locations comprising genomic regions can comprise sequences transcribed in certain tissues but not in other tissues.

[0109] The genomic locations in the panel can comprise coding and / or non-coding sequences. For example, the genomic locations in the panel can comprise one or more sequences in exons, introns, promoters, 3’ untranslated regions, 5’ untranslated regions, regulatory elements, transcription start sites, and / or splice sites. In some cases, the regions in the panel can comprise other non-coding sequences, including pseudogenes, repeat sequences, transposons, viral elements, and telomeres. In some cases, the genomic locations in the panel can comprise sequences in non-coding RNA, e.g., ribosomal RNA, transfer RNA, Piwi-interacting RNA, and microRNA.

[0110] The genomic locations in the panel can be selected to detect (diagnose) a cancer with a desired level of sensitivity (e.g., through the detection of one or more genetic variants). For example, the regions in the panel can be selected to detect the cancer (e.g., through the detection of one or more genetic variants) with a sensitivity of at least 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, 96%, 97%, 98%, 99%, 99.5%, or 99.9%. The genomic locations in the panel can be selected to detect the cancer with a sensitivity of 100%.

[0111] The genomic locations in the panel can be selected to detect (diagnose) a cancer with a desired level of specificity (e.g., through the detection of one or more genetic variants). For example, the genomic locations in the panel can be selected to detect cancer (e.g., through the detection of one or more genetic variants) with a specificity of at least 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, 96%, 97%, 98%, 99%, 99.5%, or 99.9%. The genomic locations in the panel can be selected to detect the one or more genetic variant with a specificity of 100%.

[0112] The genomic locations in the panel can be selected to detect (diagnose) a cancer with a desired positive predictive value. Positive predictive value can be increased by increasing sensitivity (e.g., chance of an actual positive being detected) and / or specificity (e.g., chance of not mistaking an actual negative for a positive). As a non-limiting example, genomic locations in the panel can be selected to detect the one or more genetic variant with a positive predictive value of at least 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, 96%, 97%, 98%, 99%, 99.5%, or 99.9%. The regions in the panel can be selected to detect the one or more genetic variant with a positive predictive value of 100%.

[0113] The genomic locations in the panel can be selected to detect (diagnose) a cancer with a desired accuracy. As used herein, the term “accuracy” may refer to the ability of a test to discriminate between a disease condition (e.g., cancer) and healthy condition.Attorney Docket No. GH0286WOAccuracy can be quantified using measures such as sensitivity and specificity, predictive values, likelihood ratios, the area under the ROC curve, Youden’s index and / or diagnostic odds ratio.

[0114] The present disclosure further provides methods for determining mutational signatures from targeted sequencing data. In various embodiments, somatic SNVs identified in targeted sequencing regions are categorized according to trinucleotide mutational contexts. For example, mutations may be classified according to the substituted base and flanking nucleotide sequence to generate 96 mutational contexts.

[0115] Normalized mutational count matrices may subsequently be generated for each sample and projected onto reference mutational signatures including COSMIC mutational signatures. In certain embodiments, projection scores are determined using nonnegative least squares methods, non-negative matrix factorization, singular value decomposition, principal component analysis, or related matrix decomposition approaches. In some embodiments, the disclosed methods identify signatures associated with homologous recombination deficiency including SBS3 and SBS39. Additional signatures associated with mismatch repair deficiency, including SBS6, SB SI 5, and SBS26, may be negatively associated with HRD and incorporated into ensemble classifiers to improve specificity. The present disclosure additionally recognizes that mutational signatures derived from wholegenome sequencing datasets may not transfer directly to targeted sequencing assays or ctDNA workflows. Accordingly, the disclosed systems utilize projection-based and stability-enhanced approaches specifically adapted for targeted sequencing environments.

[0116] In various embodiments, genomic scarring metrics are determined from copynumber profiles generated from targeted sequencing data. In certain embodiments, the genomic scarring metrics comprise loss of heterozygosity, large-scale state transitions, telomeric allelic imbalance, allelic imbalance burden, chromosomal rearrangement burden, or combinations thereof. In some embodiments, loss of heterozygosity is determined by identifying genomic regions exhibiting loss of one parental allele. Large-scale state transitions may be determined by identifying chromosomal breaks between adjacent genomic regions exceeding a threshold genomic size. Telomeric allelic imbalance may be determined by identifying allelic imbalance extending toward subtelomeric regions without crossing centromeres. The disclosed systems further recognize that polyploid tumors may artificially inflate certain genomic scarring metrics, particularly LST values. Accordingly, in certain embodiments, LST scores are normalized according to estimated tumor ploidy to improve classifier specificity. In various embodiments, tumor ploidy is estimated using copy-numberAttorney Docket No. GH0286WOsegmentation data, allele frequencies, tumor fraction estimates, and model fitting approaches. In certain embodiments, LST scores are adjusted using ploidy-specific normalization factors or generalized polyploid correction factors.

[0117] The present disclosure additionally provides quality control methods configured to improve robustness of CNV-based HRD prediction. In certain embodiments, the disclosed systems compute quality metrics including mean CN distance and mean MAF distance to assess concordance between observed genomic measurements and model-predicted genomic states.

[0118] Samples exhibiting unstable ploidy estimates, noisy copy -number segmentation, poor CNV fit, or excessive modeling residuals may be classified as not evaluable. Such approaches reduce false-positive HRD calls and improve reproducibility.

[0119] In an implementation, after sequencing, sequence reads may be assigned a quality score. A quality score may be a representation of sequence reads that indicates whether those sequence reads may be useful in subsequent analysis based on a threshold. In some cases, some sequence reads are not of sufficient quality or length to perform a subsequent mapping step. Sequence reads with a quality score at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of a data set of sequence reads. In other cases, sequence reads assigned a quality scored at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. Sequence reads that meet a specified quality score threshold may be mapped to a reference genome. After mapping alignment, sequence reads may be assigned a mapping score. A mapping score may be a representation of sequence reads mapped back to the reference sequence indicating whether each position is or is not uniquely mappable. Sequence reads with a mapping score at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. In other cases, sequencing reads assigned a mapping scored less than 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set.Precision Treatments

[0120] The precision diagnostics provided by the improved computer system 110 may result in precision treatment plans, which may be identified by the computer system 110 (and / or curated by health professionals). For example, one type of precision diagnostic and treatment may relate to genes in the homologous recombination repair (HRR) pathway.Attorney Docket No. GH0286WO

[0121] Homologous recombination is a type of genetic recombination in which nucleotide sequences are exchanged between two similar or identical molecules of DNA. It is most widely used by cells to accurately repair harmful breaks that occur on both strands of DNA, known as double-strand breaks (DSB). HRR provides a mechanism for the error-free removal of damage present in DNA that has replicated (S and G2 phases), to eliminate chromosomal breaks before the cell division occurs. The primary model for how homologous recombination repairs double-strand breaks in DNA is homologous recombination repair pathway which mediates the double-strand break repair (DSBR) pathway and the synthesisdependent strand annealing (SDSA) pathway. Germline and somatic deficiencies in homologous recombination genes have been strongly linked to breast, ovarian and prostate cancers.

[0122] The number and types of variant nucleotides in a sample can provide an indication of the amenability of the subject providing the sample to treatment, i.e., therapeutic intervention. For example, various poly ADP ribose polymerase (PARP) inhibitors have been shown to stop the growth of tumors from breast, ovarian and prostate cancers caused by hereditary mutations in the BRCA1 or BRCA2 genes. Some of these therapeutic agents may inhibit base excision repair (BER), which may compensate for the deficiency of HRR.

[0123] On the other hand, certain BRCA and HRR wildtype patients may not achieve clinical benefit from treatment with a PARP inhibitor. Furthermore, not all ovarian cancer patients with a BRCA mutation will respond to a PARP inhibitor. Moreover, different types of mutations may indicate different therapies. For example, somatic heterozygous deletions in HRR genes may indicate a different therapy than somatic homozygous deletions. Thus, the state of genetic material may influence therapy. In one example, a PARP inhibitor may be administered to an individual harboring a somatic homozygous deletion in a HRR gene, but not to an individual harboring a wildtype allele or somatic heterozygous deletions in the HRR gene.

[0124] In some implementations, a subject having HRD as determined by any of the methods disclosed may be administered a targeted therapy. The targeted therapy may comprise a PARP inhibitor. Examples of PARP inhibitors that may be administered include one or more of: VELIPARIB, OLAPARIB, TALAZOPARIB, RUC APARIB, NIRAPARIB, PAMIPARIB, CEP 9722 (Cephalon), E7016 (Eisai), E7449 (Eisai, a PARP 1 / 2 and tankyrase 1 / 2 inhibitor), or 3 -Aminobenzamide. In some implementations, the targeted therapy may comprise at least one base excision repair (BER) inhibitor. For example, OLAPARIB may inhibit BER. In certain implementations, the targeted therapy may comprise combination of aAttorney Docket No. GH0286WOPARP inhibitor and radiotherapy. In an implementation, the combination of a PARP inhibitor and radiotherapy would permit the PARP inhibitor to lead to formation of double strand breaks from the single-strand breaks generated by the radiotherapy in tumor tissue (e.g., tissue with BRCA1 / BRCA2 mutations). This combination can provide more powerful therapy per radiation dose.Customized Therapies and Related Administrations

[0125] In some implementations, the methods disclosed herein relate to identifying and administering therapies to patients having a given disease, disorder or condition.Essentially any cancer therapy (e.g., surgical therapy, radiation therapy, chemotherapy, and / or the like) is included as part of these methods. In certain implementations, the therapy administered to a subject may comprise at least one chemotherapy drug. In some implementations, the chemotherapy drug may comprise alkylating agents (for example, but not limited to, Chlorambucil, Cyclophosphamide, Cisplatin and Carboplatin), nitrosoureas (for example, but not limited to, Carmustine and Lomustine), anti-metabolites (for example, but not limited to, Fluorauracil, Methotrexate and Fludarabine), plant alkaloids and natural products (for example, but not limited to, Vincristine, Paclitaxel and Topotecan), anti- tumor antibiotics (for example, but not limited to, Bleomycin, Doxorubicin and Mitoxantrone), hormonal agents (for example, but not limited to, Prednisone, Dexamethasone, Tamoxifen and Leuprolide) and biological response modifiers (for example, but not limited to, Herceptin and Avastin, Erbitux and Rituxan). In some implementations, the chemotherapy administered to a subject may comprise FOLFOX or FOLFIRI. Typically, therapies include at least one immunotherapy (or an immunotherapeutic agent). Immunotherapy refers generally to methods of enhancing an immune response against a given cancer type. In certain implementations, immunotherapy refers to methods of enhancing a T cell response against a tumor or cancer.

[0126] In some implementations, the immunotherapy or immunotherapeutic agents targets an immune checkpoint molecule. Certain tumors are able to evade the immune system by co-opting an immune checkpoint pathway. Thus, targeting immune checkpoints has emerged as an effective approach for countering a tumor’s ability to evade the immune system and activating anti-tumor immunity against certain cancers. Pardoll, Nature Reviews Cancer, 2012, 12:252-264.Attorney Docket No. GH0286WO

[0127] In certain implementations, the immune checkpoint molecule is an inhibitory molecule that reduces a signal involved in the T cell response to antigen. For example, CTLA4 is expressed on T cells and plays a role in downregulating T cell activation by binding to CD80 (aka B7.1) or CD86 (aka B7.2) on antigen presenting cells. PD-1 is another inhibitory checkpoint molecule that is expressed on T cells. PD-1 limits the activity of T cells in peripheral tissues during an inflammatory response. In addition, the ligand for PD-1 (PD-L1 or PD-L2) is commonly upregulated on the surface of many different tumors, resulting in the downregulation of anti-tumor immune responses in the tumor microenvironment. In certain implementations, the inhibitory immune checkpoint molecule is CTLA4 or PD-1. In other implementations, the inhibitory immune checkpoint molecule is a ligand for PD-1, such as PD-L1 or PD-L2. In other implementations, the inhibitory immune checkpoint molecule is a ligand for CTLA4, such as CD80 or CD86. In other implementations, the inhibitory immune checkpoint molecule is lymphocyte activation gene 3 (LAG3), killer cell immunoglobulin like receptor (KIR), T cell membrane protein 3 (TIM3), galectin 9 (GAL9), or adenosine A2a receptor (A2aR).

[0128] Antagonists that target these immune checkpoint molecules can be used to enhance antigen-specific T cell responses against certain cancers. Accordingly, in certain implementations, the immunotherapy or immunotherapeutic agent is an antagonist of an inhibitory immune checkpoint molecule. In certain implementations, the inhibitory immune checkpoint molecule is PD-1. In certain implementations, the inhibitory immune checkpoint molecule is PD-L1. In certain implementations, the antagonist of the inhibitory immune checkpoint molecule is an antibody (e.g., a monoclonal antibody). In certain implementations, the antibody or monoclonal antibody is an anti-CTLA4, anti-PD-1, anti-PD-L1, or anti-PD-L2 antibody. In certain implementations, the antibody is a monoclonal anti-PD-1 antibody. In some implementations, the antibody is a monoclonal anti-PD-Ll antibody. In certain implementations, the monoclonal antibody is a combination of an anti-CTLA4 antibody and an anti-PD-1 antibody, an anti-CTLA4 antibody and an anti-PD-Ll antibody, or an anti-PD-Ll antibody and an anti-PD-1 antibody. In certain implementations, the anti-PD-1 antibody is one or more of pembrolizumab (Keytruda®) or nivolumab (Opdivo®). In certain implementations, the anti-CTLA4 antibody is ipilimumab (Yervoy®). In certain implementations, the anti-PD-Ll antibody is one or more of atezolizumab (Tecentriq®), avelumab (Bavencio®), or durvalumab (Imfinzi®), anti-VEGF(Bevacizumab®).Attorney Docket No. GH0286WO

[0129] In certain implementations, the immunotherapy or immunotherapeutic agent is an antagonist (e.g., antibody) against CD80, CD86, LAG3, KIR, TIM3, GAL9, or A2aR. In other implementations, the antagonist is a soluble version of the inhibitory immune checkpoint molecule, such as a soluble fusion protein comprising the extracellular domain of the inhibitory immune checkpoint molecule and an Fc domain of an antibody. In certain implementations, the soluble fusion protein comprises the extracellular domain of CTLA4, PD-1, PD-L1, or PD-L2. In some implementations, the soluble fusion protein comprises the extracellular domain of CD80, CD86, LAG3, KIR, TIM3, GAL9, or A2aR. In one implementation, the soluble fusion protein comprises the extracellular domain of PD-L2 or LAG3.

[0130] In certain implementations, the immune checkpoint molecule is a costimulatory molecule that amplifies a signal involved in a T cell response to an antigen. For example, CD28 is a co-stimulatory receptor expressed on T cells. When a T cell binds to antigen through its T cell receptor, CD28 binds to CD80 (aka B7.1) or CD86 (aka B7.2) on antigen-presenting cells to amplify T cell receptor signaling and promote T cell activation. Because CD28 binds to the same ligands (CD80 and CD86) as CTLA4, CTLA4 is able to counteract or regulate the co-stimulatory signaling mediated by CD28. In certain implementations, the immune checkpoint molecule is a co-stimulatory molecule selected from CD28, inducible T cell co-stimulator (ICOS), CD137, 0X40, or CD27. In other implementations, the immune checkpoint molecule is a ligand of a co-stimulatory molecule, including, for example, CD80, CD86, B7RP1, B7-H3, B7-H4, CD137L, OX40L, or CD70.

[0131] Agonists that target these co-stimulatory checkpoint molecules can be used to enhance antigen-specific T cell responses against certain cancers. Accordingly, in certain implementations, the immunotherapy or immunotherapeutic agent is an agonist of a co-stimulatory checkpoint molecule. In certain implementations, the agonist of the co-stimulatory checkpoint molecule is an agonist antibody and preferably is a monoclonal antibody. In certain implementations, the agonist antibody or monoclonal antibody is an anti-CD28 antibody. In other implementations, the agonist antibody or monoclonal antibody is an anti-ICOS, anti-CD137, anti-OX40, or anti-CD27 antibody. In other implementations, the agonist antibody or monoclonal antibody is an anti-CD80, anti-CD86, anti-B7RPl, anti-B7-H3, anti-B7-H4, anti-CD137L, anti-OX40L, or anti-CD70 antibody.

[0132] Therapeutic options for treating specific genetic-based diseases, disorders, or conditions, other than cancer, are generally well-known to those of ordinary skill in the art and will be apparent given the particular disease, disorder, or condition under consideration.Attorney Docket No. GH0286WO

[0133] In certain implementations, the customized therapies described herein are typically administered parenterally (e.g., intravenously or subcutaneously). Pharmaceutical compositions containing the immunotherapeutic agent are typically administered intravenously. Certain therapeutic agents are administered orally. However, customized therapies (e.g., immunotherapeutic agents, etc.) may also be administered by any method known in the art, including, for example, buccal, sublingual, rectal, vaginal, intraurethral, topical, intraocular, intranasal, and / or intraauricular, which administration may include tablets, capsules, granules, aqueous suspensions, gels, sprays, suppositories, salves, ointments, or the like.

[0134] The systems and methods disclosed herein may be used to identify patients likely to benefit from DNA damage response-targeted therapies. In certain embodiments, HRD-positive subjects are selected for treatment with PARP inhibitors including olaparib, rucaparib, niraparib, and talazoparib.

[0135] In some embodiments, HRD classification predicts responsiveness to platinum chemotherapy, ATR inhibitors, DNA-PK inhibitors, immune checkpoint inhibitors, or combinations thereof. The disclosed methods may additionally be used to predict clinical outcomes including progression-free survival, radiographic progression-free survival, overall survival, and time to PSA progression. The present disclosure further provides methods for identifying HRD-positive tumors lacking pathogenic BRCA1 or BRCA2 mutations, thereby expanding therapeutic eligibility beyond conventional mutation-based testing approaches.Computer Systems

[0136] Methods of the present disclosure can be implemented using, or with the aid of, computer systems. For example, such methods may comprise (a) determining a plurality of quantitative measures for the nucleic acid variant from a sample of nucleic acid molecules (e.g., a sample of cfDNA), wherein the plurality of quantitative measures comprises a total allele count and a minor allele count for the nucleic acid variant; (b) identifying an associated variable of the nucleic acid variant from the sample; (c) determining a quantitative value for the associated variable of the nucleic acid variant; (d) generating a statistical model for expected germline mutant allele counts at a genomic locus of the nucleic acid variant; (e) generating a probability value (p-value) for the nucleic acid variant based at least in part on the statistical model for expected germline mutant allele counts, the quantitative value for the associated variable of the nucleic acid variant, and at least one of the plurality of quantitative measures for the nucleic acid variant; and (f) classifying the nucleic acid variant as (i) beingAttorney Docket No. GH0286WOof somatic origin when the p-value for the nucleic acid variant is below a predetermined threshold value, or as (ii) being of germline origin when the p-value for the nucleic acid variant is at or above the predetermined threshold value, can be performed with a computer processor.

[0137] Additional details relating to computer systems and networks, databases, and computer program products are also provided in, for example, Peterson, Computer Networks: A Systems Approach, Morgan Kaufmann, 5th Ed. (2011), Kurose, Computer Networking: A Top-Down Approach, Pearson, 7th Ed. (2016), Elmasri, Fundamentals of Database Systems, Addison Wesley, 6th Ed. (2010), Coronel, Database Systems: Design, Implementation, & Management, Cengage Learning, 11th Ed. (2014), Tucker, Programming Languages, McGraw-Hill Science / Engineering / Math, 2nd Ed. (2006), and Rhoton, Cloud Computing Architected: Solution Design Handbook, Recursive Press (2011), each of which is hereby incorporated by reference in its entirety.Machine Learning

[0138] The methods and systems herein can be applied to a classifier. The classifier can be trained or untrained. The classifier is used to identify patterns associated with a condition or state of a condition. A classifier may be implemented on a computer.

[0139] In as aspect, a classifier may determine genetic aberrations in a test subject using DNA from a cell-free sample (or cell-free DNA) obtained from the test subject. This classifier may comprise (a) an input of a set of distribution scores for each of one or more samples (or cell-free DNA) from subjects, wherein each distribution score is representative of a number of bases present in DNA from a cell-free sample (or cell-free DNA) from a subject that map to each of a plurality of positions in a genome; and (b) an output of classifications of one or more genetic aberrations.

[0140] A classifier may comprise a machine learning engine. The distribution scores may represent length of each molecule from which a base position is mapped. The distribution scores may represent counts of each molecule overlapping a base position. The distribution scores may represent counts of each molecule starting at a base position. The distribution scores may represent counts of each molecule ending at a base position.

[0141] A classifier may be used to determine genetic aberrations in a test subject using DNA from a cell-free sample (or cell-free DNA) obtained from the test subject by providing a set of distribution scores for a test subject, and generating a classification of the test subject using the classifier.Attorney Docket No. GH0286WO

[0142] A classifier may be trained by a training set. A training set may comprise a set of distribution scores for each of a plurality of samples from subjects and a set of classifications for each of the plurality of samples. The set of distribution scores may comprise (a) a set of reference distribution scores for each of a plurality of samples from control subjects, wherein each reference distribution score is representative of a number of bases present in DNA from a cell-free sample (or cell-free DNA) from a control subject that map to each of a plurality of positions in a genome or (b) a set of phenotypic distribution scores for each of a plurality of samples from subjects having an observed phenotype, wherein each phenotypic distribution score is representative of a number of bases present in DNA from a cell-free sample (or cell-free DNA) from a subject having the observed phenotype that map to each of a plurality of positions in a genome. The set of classifications may comprise (c) a set of reference classifications for each of the plurality of samples from control subjects or (d) a set of phenotypic classifications for each of the plurality of samples from subjects having an observed phenotype.

[0143] The control subjects associated with the set of reference distribution scores or the set of reference classifications may be asymptomatic healthy individuals. The subjects having an observed phenotype associated with the set of phenotypic distribution scores or the set of phenotypic classifications may comprise (a) subjects with a tissue-specific cancer, (b) subjects with a particular stage of cancer, (c) subjects with an inflammatory condition, (d) subjects that are asymptomatic to cancer but have a tumor that will progress into cancer, or (e) subjects with cancer having positive or negative response to a particular drug or drug regimen.

[0144] The classifier may further comprise an input of a set of genetic variants at one or more loci of the genome. The set of genetic variants may comprises one or more loci of reported tumor markers (e.g., a reported tumor marker in COSMIC).

[0145] A method is provided for creating a trained classifier, comprising (a) providing a plurality of different classes, wherein each class represents a set of subjects with a shared characteristic (e.g., from one or more cohorts); (b) providing a uni-parametric or multi-parametric model representative of the cell-free DNA molecules from each of a plurality of samples belonging to each of the classes, thereby providing a training data set; and (c) training a learning algorithm on the training data set to create one or more trained classifiers, wherein each trained classifier classifies a test sample into one or more of the plurality of classes.Attorney Docket No. GH0286WO

[0146] As an example, a trained classifier may use a learning algorithm selected from the group consisting of a random forest, a neural network, a support vector machine, and a linear classifier. Each of the plurality of different classes may be selected from the group consisting of healthy, breast cancer, colon cancer, lung cancer, pancreatic cancer, prostate cancer, ovarian cancer, melanoma, and liver cancer.

[0147] A trained classifier may be applied to a method of classifying a sample from a subject. This method of classifying may comprise: (a) providing a set of one or more uniparametric models representative of the cell-free DNA molecules from a test sample from the subject; and (b) classifying the test sample using a trained classifier. After the test sample is classified into one or more classes, performing a therapeutic intervention on the subject based on the classification of the sample.

[0148] A trained classifier may be applied to a method of classifying a sample from a subject. This method of classifying may comprise: (a) providing a multi -parametric model representative of the cell-free DNA molecules from a test sample from the subject; and (b) classifying the test sample using a trained classifier. After the test sample is classified into one or more classes, performing a therapeutic intervention on the subject based on the classification of the sample.EXAMPLESExample 1 — HRD Score, Generally

[0149] Described herein are methods and compositions related to use of models, including probabilistic models, ensemble models and the like, in evaluating samples for HRD. Specifically, described herein are systems and methods for determining homologous recombination deficiency (HRD), genomic instability status (GIS), and responsiveness to DNA damage response (DDR)-targeted therapies using targeted sequencing data derived from biological samples including tissue and liquid biopsy samples.

[0150] In particular embodiments, the disclosure relates to probabilistic and ensemble-based machine learning approaches utilizing single nucleotide variant (SNV) mutational signatures, copy -number variation (CNV) features, genomic scarring signatures, epigenomic biomarkers, transcriptomic biomarkers, and quality control metrics to classify tumors as HRD-positive or HRD-negative.Attorney Docket No. GH0286WOExample 2 - Generation of Mutational Signature Projection Scores

[0151] Targeted sequencing data derived from breast, prostate, ovarian, and pancreatic cancer clinical samples were processed to identify somatic SNVs present within targeted genomic regions. For each identified SNV, trinucleotide mutational contexts were determined by evaluating the substituted nucleotide together with immediately adjacent upstream and downstream nucleotides. Mutation counts corresponding to all 96 trinucleotide contexts were subsequently compiled into mutational count matrices representing individual tumor samples.

[0152] The mutational count matrices were normalized to account for sequencing depth and variable mutation burden between samples. The normalized matrices were then projected onto COSMIC mutational signatures version 3.4 using non-negative least squares decomposition methods. Projection scores corresponding to individual mutational signatures were calculated for each sample. Statistical comparisons between HRD-positive samples and HRR- wildtype samples demonstrated significant enrichment of HRD-associated signatures including SBS3 and SBS39 in HRD-positive samples. In contrast, mismatch repair deficiency-associated signatures including SBS6 and SBS15 exhibited reduced projection scores in HRD-positive tumors.

[0153] The projection-based approach demonstrated improved stability relative to earlier de novo signature derivation methods. Because the disclosed methods utilized predefined biologically relevant signatures rather than retraining unstable de novo signatures, replicate concordance substantially improved in low-input liquid biopsy samples.Example 3 — Ensemble GIS Classifier

[0154] An ensemble logistic regression classifier was generated using combined genomic scarring and mutational signature features derived from targeted sequencing assays. Training cohorts included breast, prostate, ovarian, and pancreatic clinical samples processed using targeted next-generation sequencing workflows. HRD-positive labels were assigned to samples exhibiting biallelic BRCA1 or BRCA2 loss, while HRR-wildtype labels were assigned to samples lacking deleterious HRR mutations, promoter methylation, loss of heterozygosity, or pathogenic germline variants across a predefined HRR gene set.Attorney Docket No. GH0286WO

[0155] The classifier utilized copy-number features including LST, LOH, and TAI together with mutational signature projection scores. Samples containing at least 20 somatic SNVs were analyzed using a combined CNV+SNV model, while samples containing fewer than 20 SNVs utilized a CNV-only classifier. Probabilistic GIS scores ranging from 0 to 1 were generated for each sample.

[0156] Independent test cohorts demonstrated strong classifier performance across multiple tumor types. In particular embodiments, pan-cancer ROC AUC values approached or exceeded 0.94-0.98 depending on cohort composition and tumor fraction thresholds. The ensemble approach further improved sensitivity in breast cancer samples while maintaining specificity in prostate cancer samples.Example 4 — Ploidy-Normalized LST Scores

[0157] Copy -number segmentation profiles from polyploid tumors were analyzed to evaluate the effect of ploidy on genomic scarring metrics. Raw LST scores demonstrated inflation in highly polyploid samples, resulting in increased false-positive HRD classifications in certain low-HRD tumor types including non-small cell lung cancer and head and neck cancer.

[0158] To address this issue, tumor ploidy was estimated from allele-specific copynumber profiles and tumor fraction measurements. LST values were subsequently normalized using ploidy-based correction factors derived from training cohorts. In some implementations, distinct correction factors were utilized for different ploidy ranges. In alternative implementations, generalized polyploid normalization factors were applied.

[0159] Ploidy -normalized LST values improved classifier specificity and reduced false-positive HRD calls associated with highly rearranged tetrapioid tumors. In certain cohorts, ploidy normalization improved separation between HRD-positive and HRR-wildtype samples and reduced instability associated with noisy ploidy estimation.Example 5 — CNV Quality Control Filtering

[0160] Copy -number model fitting performance was evaluated using quality control metrics configured to assess concordance between observed genomic measurements and predicted genomic states. Mean CN distance and mean MAF distance were calculated across segmented genomic regions to quantify model fit.Attorney Docket No. GH0286WO

[0161] Samples exhibiting excessive discordance between observed and predicted copy-number values demonstrated elevated false-positive HRD classifications, particularly in tumors exhibiting noisy CNV profiles. Threshold values for mean CN distance and mean MAF distance were therefore established based on empirical distributions observed in training and validation cohorts.

[0162] Samples exceeding predefined QC thresholds were classified as not evaluable. Application of these quality control approaches substantially reduced false-positive HRD calls while preserving sensitivity in clinically relevant HRD-positive tumors.Example 6 — HRD Detection Beyond BRCA1 / 2 Mutations

[0163] Clinical tissue and liquid biopsy cohorts were evaluated to determine the prevalence of GIS-detected tumors lacking pathogenic BRCA1 / 2 mutations. Across breast, ovarian, pancreatic, and prostate cancer samples, a substantial proportion of GIS-positive tumors lacked canonical BRCA1 or BRCA2 pathogenic variants.

[0164] In breast cancer cohorts, GIS positivity exceeded the prevalence of BRC Allmutated tumors, demonstrating expansion of HRD detection beyond mutation-only approaches. Similar findings were observed in ovarian and pancreatic cancer cohorts. A subset of GIS-positive tumors harbored pathogenic alterations in additional HRR genes including PALB2, RAD51C, and RAD51D, while other GIS-positive tumors lacked clearly pathogenic HRR mutations altogether.

[0165] These findings demonstrated that the disclosed ensemble GIS methods identify clinically relevant HRD biology extending beyond conventional BRCA mutation testing and may expand patient populations eligible for DDR-targeted therapies.Example 7 — Clinical Association with PARP Inhibitor Response

[0166] Pre-treatment plasma samples obtained from metastatic castration-resistant prostate cancer patients enrolled in a PARP inhibitor clinical trial were analyzed using the disclosed GIS classifier. Patients included individuals harboring BRCA1, BRCA2, PALB2, FANCA, CDK12, and related HRR alterations.

[0167] GIS-positive status was significantly associated with improved progression-free survival and radiographic progression-free survival. Cox proportional hazards analyses demonstrated reduced hazard ratios for progression among GIS-positive patients afterAttorney Docket No. GH0286WOadjustment for age and tumor fraction covariates. In certain analyses, GIS status provided predictive value beyond BRCA or PALB2 mutation status alone.

[0168] Use of prostate-specific thresholds and CNV quality control filtering further improved classifier specificity by reducing false-positive calls associated with noisy LOH profiles and polyploid tumors. These findings demonstrated that the disclosed systems provide clinically actionable prediction of PARP inhibitor responsiveness.Example 8 — Tissue-Based GIS Prediction

[0169] A targeted tissue sequencing assay configured to profile approximately 741 genes was utilized to generate probabilistic GIS scores from FFPE tumor samples. The assay simultaneously detected SNVs, indels, copy-number alterations, gene fusions, methylation biomarkers, and transcriptomic biomarkers.

[0170] Training and validation cohorts included breast, ovarian, and pancreatic clinical samples. HRD-positive labels were assigned based on biallelic BRCA1 or BRCA2 loss, while HRR-wildtype labels were assigned using strict exclusion criteria across multiple HRR genes. Mutational signature projection scores and genomic scarring metrics were integrated into ensemble classifiers.

[0171] The resulting tissue-based GIS classifier achieved high sensitivity and specificity across multiple cancer types. In certain cohorts, sensitivity exceeded 88-96% in ovarian cancer and approximately 78-96% in breast cancer. The classifier further identified substantial populations of GIS-positive tumors lacking pathogenic BRCA1 or BRCA2 mutations.Example 9 — Dual Tissue and Liquid Monitoring

[0172] Matched tissue-derived and plasma-derived nucleic acid samples obtained from cancer patients were analyzed longitudinally to evaluate coordinated GIS determination across tissue and liquid biopsy assays. Tissue-derived genomic instability measurements demonstrated strong concordance with liquid biopsy-derived genomic instability measurements in evaluable samples.

[0173] Longitudinal monitoring enabled assessment of HRD dynamics during treatment, including emergence of resistance, changes in tumor fraction, and therapy-associated genomic evolution. In some embodiments, tissue-derived GIS measurements wereAttorney Docket No. GH0286WOused to establish baseline HRD status while liquid biopsy-derived GIS measurements were used to monitor treatment response and disease progression over time.

[0174] The disclosed dual tissue-liquid workflow demonstrated that harmonized classifier architectures and genomic instability frameworks can be applied across multiple sample types to provide clinically useful longitudinal HRD assessment.

Claims

Attorney Docket No. GH0286WOCLAIMS1. A computer-implemented method for determining homologous recombination deficiency (HRD) status in a biological sample, comprising:obtaining sequencing data from a targeted next-generation sequencing assay; determining one or more single nucleotide variant (SNV) mutations from the sequencing data;generating, using a classifier, a probabilistic HRD score based on one or more of: SNV mutations, CNV features, loss of heterozygosity (LOH),large-scale state transitions (LST), telomeric allelic imbalance (TAI), allelic imbalance, and copy -number burden; andclassifying the biological sample as HRD-positive or HRD-negative based on the probabilistic HRD score.

2. The method of claim 1, determining one or more copy -number variation (CNV) features from the sequencing data; wherein the CNV features comprise one or more of: loss of heterozygosity (LOH), large-scale state transitions (LST), telomeric allelic imbalance (TAI), allelic imbalance, and copy-number burden.

3. The method of claim 1, wherein one or more mutational signatures comprise one or more SNV mutations and / or COSMIC signatures.

4. The method of claim 3, wherein the COSMIC signatures comprise SBS3.

5. The method of claim 3, wherein the COSMIC signatures comprise SBS3, SBS39, SBS6, SBS15, SBS26, or any combination thereof.

6. The method of claim 1, wherein the classifier comprises a logistic regression model.

7. The method of claim 1, wherein the classifier comprises an ensemble classifier.

8. The method of claim 7, wherein the ensemble classifier selectively applies a CNV- only model and / or a CNV+SNV model.Attorney Docket No. GH0286WO9. The method of claim 8, wherein model selection depends on a quantity of detected somatic SNVs.

10. The method of claim 9, wherein the CNV+SNV model is used when at least 10, 15, 20, 25 or more, optionally 20, somatic SNVs are detected.

11. The method of claim 1, further comprising estimating tumor ploidy.

12. The method of claim 11, further comprising adjusting one or more LST values based on the tumor ploidy.

13. The method of claim 1, further comprising determining one or more CNV quality control metrics.

14. The method of claim 13, wherein the CNV quality control metrics comprise mean copy (CN) distance.

15. The method of claim 13, wherein the CNV quality control metrics comprise mean mutant allele fraction (MAF) distance.

16. The method of claim 13, wherein samples exceeding a threshold CNV quality control value are classified as not evaluable.

17. The method of claim 1, wherein the biological sample comprises a liquid and / or tissue biopsy sample.

18. The method of claim 17, wherein the liquid biopsy sample comprises circulating tumor and / or cell-free DNA.

19. The method of claim 1, wherein the biological sample comprises a tissue sample.

20. The method of claim 1, wherein the targeted sequencing assay interrogates at least 500 genes.Attorney Docket No. GH0286WO21. The method of claim 20, wherein the targeted sequencing assay interrogates 600-800, optionally at least 700, genes.

22. The method of claim 1, wherein the sequencing data further comprises methylation biomarkers.

23. The method of claim 1, wherein the sequencing data further comprises transcriptomic biomarkers.

24. The method of claim 1, wherein the classifier further utilizes epigenomic biomarkers.

25. The method of claim 1, wherein the classifier further utilizes transcriptomic biomarkers.

26. The method of claim 1, wherein the probabilistic HRD score ranges from 0 to 1.

27. The method of claim 1, wherein the biological sample is classified as HRD detected, HRD not detected, or HRD not evaluable.

28. The method of claim 1, further comprising selecting a therapy based on the HRD classification.

29. The method of claim 28, wherein the therapy comprises a PARP and / or BER inhibitor.

30. The method of claim 29, wherein the PARP inhibitor comprises olaparib, rucaparib, niraparib, or talazoparib.

31. The method of claim 28, wherein the therapy comprises platinum chemotherapy.

32. The method of claim 28, wherein the therapy comprises an ATR inhibitor.

33. The method of claim 1, wherein the method predicts responsiveness to DNA damage response therapy.Attorney Docket No. GH0286WO34. The method of claim 1, wherein the method predicts progression-free survival.

35. The method of claim 1, wherein the method predicts radiographic progression-free survival.

36. The method of claim 1, wherein the method identifies HRD-positive tumors lacking pathogenic BRCA1 / 2 mutations.

37. The method of claim 1, wherein HRD-positive training samples comprise samples exhibiting biallelic BRCA1 or BRCA2 loss.

38. The method of claim 1, wherein HRD-negative training samples comprise samples lacking deleterious HRR mutations, promoter methylation in HRR genes, loss of heterozygosity in HRR genes, and / or pathogenic germline HRR variants.

39. The method of claim 1, wherein comprising determining mutational signatures using normalized mutational count matrices.

40. The method of claim 39, wherein normalized mutational count matrices comprise 96 trinucleotide contexts.

41. The method of claim 1, wherein the method further comprises generating mutational signature projection scores using reference mutational signatures.

42. The method of claim 41, wherein the projection scores are generated by matrix decomposition.

43. The method of claim 41, wherein the projection scores are generated using nonnegative least squares.

44. A system configured to perform the method of claim 1.Attorney Docket No. GH0286WO45. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform the method of claim 1.

46. A kit for determining homologous recombination deficiency (HRD), comprising: reagents for targeted next-generation sequencing, and / or software configured to generate a probabilistic HRD score using SNV mutational signatures and CNV features.

47. The kit of claim 46, wherein the software applies a ploidy-adjusted LST score.

48. The kit of claim 46, wherein the software classifies samples as HRD detected, HRD not detected, or HRD not evaluable.

49. A method of monitoring HRD status in a subject, comprising:determining a first HRD score from a first biological sample obtained at a first time point; anddetermining a second HRD score from a second biological sample obtained at a second time point.

50. The method of claim 49, wherein one sample comprises tissue-derived nucleic acids and another sample comprises circulating tumor DNA.