Abstract Background: Binary DNA damage repair (DDR) classifications fail to capture tumor complexity. Homologous recombination repair (HRR) and mismatch repair (MMR) testing overlooks concurrent deficiencies, clonal heterogeneity, and repair-intact phenotypes, causing therapeutic misclassification for PARP inhibitor (PARPi) and immunotherapy. We hypothesized probabilistic DDR classification integrating genomic and transcriptomic features would enable precision stratification. Methods: We analyzed 672 metastatic prostate cancers via whole-exome sequencing, mutational signatures, and copy number profiling. Transcriptomic analysis of 662 tumors included expression signatures and pathway analysis. Molecular characterization identified 11 subgroups based on DDR alterations, including concurrent deficiencies. We developed CHIMERA-DDR, a nested Random Forest classifier integrating 51 genomic and transcriptomic features predicting probability scores across 7 DDR subtypes. Optimal tumor mutational burden (TMB) cutoff (18. 61 mutations/Mb) was identified via logistic regression with cross-validation. Validation included leave-one-out cross-validation, independent cohort (n=37), cellularity analysis, and 130 patients with immunotherapy outcomes. Large language models assisted with language refinement. Results: Molecular analysis identified 11 DDR subgroups, revealing dichotomy within high-TMB HRR-deficient tumors. We identified TMB-very high (TMB-VH) (≥18. 61 mutations/Mb) with preserved genomic integrity (lower ploidy, p=0. 008; reduced percent genome alteration, p=0. 01), absent traditional DDR signatures, enhanced immunogenicity—distinct from conventional TMB-high. TMB-VH tumors showed highest CD8+ infiltration and superior immunotherapy response. Critically, 2. 3% exhibited concurrent TMB-high and HRR-deficient phenotypes; 1. 5% had MMR biallelic loss without mismatch repair deficiency (MMRd) signatures, and some MMR mutants with low TMB scored DDR-intact. Genotype-phenotype discordance emerged: 77. 6% of HRR-mutant tumors exhibited HRRd characteristics while 22. 4% did not; 24% displayed MMRd-predominant phenotypes despite HRR alterations, revealing pathway dominance over dual vulnerabilities. CHIMERA-DDR assigned probability scores across 7 DDR subtypes (AUC 0. 919-0. 999), resolved admixed phenotypes, and quantified pathway dominance in concurrent deficiencies. Cross-site validation (52 samples, 24 patients) confirmed stable DDR phenotypes. Conclusions: CHIMERA-DDR captures clonal architecture and DDR heterogeneity, identifying repair-intact phenotypes unsuitable for DNA-damaging monotherapy and quantifying pathway dominance. This enables precision stratification: PARPi monotherapy for HRRd-dominant tumors, ICI monotherapy for TMB-VH, or pathway-specific therapy for concurrent deficiencies. CHIMERA-DDR identifies patients likely to benefit while avoiding futile therapies, addressing unmet needs in precision oncology. Prospective trials are needed to validate CHIMERA-predicted phenotypes, with applications extending beyond prostate cancer. Citation Format: Navonil De Sarkar, Komal Sharma, Divin Wilson, Yu Wang, Paul Auer. Making invisible tumor truths visible: CHIMERA-DDR score unmasks clonal complexity and concurrent DNA repair deficiencies in advanced prostate cancer abstract. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86 (2Suppl): Abstract nr A014.
Sarkar et al. (Tue,) studied this question.