Homologous recombination deficiency (HRD) has emerged as a critical prognostic and predictive biomarker in oncology. However, current testing methods, especially those reliant on targeted panels, are plagued by inconsistent results from the same samples. This highlights the urgent need for standardized benchmarks to evaluate HRD assay performance. In phases IIa and IIb of the Chinese HRD Harmonization Project, we developed ten pairs of well-characterized DNA reference materials derived from lung, breast, and melanoma cancer cell lines and their matched normal cell lines, keeping each paired with seven cancer-to-normal mass ratios. Reference datasets for allele-specific copy number variations (ASCNVs) and HRD scores were established and validated using three sequencing methods and nine analytical pipelines. The genomic instability scores (GISs) of the reference materials ranged from 11 to 96, enabling validation across various thresholds. The ASCNV reference datasets covered a genomic span of 2340 to 2749 Mb, equivalent to 81.2% to 95.4% of the autosomes in the 37d5 reference genome. These benchmarks were subsequently utilized to assess the accuracy and reproducibility of four HRD panel assays, revealing significant variability in both ASCNV detection and HRD scores. The concordance between panel-detected GISs and reference GISs ranged from 0.81 to 0.94, with only two assays exhibiting high overall agreement with Myriad MyChoice CDx for HRD classification. This study also identified specific challenges in ASCNV detection in HRD-related regions and the profound impact of high ploidy on consistency. The established HRD reference materials and datasets provide a robust toolkit for objective evaluation of HRD testing.

Traditional approaches for capturing genomic alterations and DNA methylation require separate assays, complicating clinical workflows and limiting sample utilization, particularly with low-input materials like cell-free DNA. To address these challenges, we introduce a streamlined approach combining mutation and methylation profiling via mutation-protective strand synthesis with modified deoxycytidine triphosphates, demonstrating high concordance with standard enzymatic methyl-seq and DNA-seq in both whole-genome sequencing of cell lines and targeted sequencing of clinical samples. In potential clinical contexts, incorporating multi-omics information with this approach modestly improve circulating tumor DNA (ctDNA) detection by ~12% in pre-treatment lung cancer patients ( N = 26) while preserving specificity in healthy controls ( N = 13), and reveal relationships between homologous recombination repair (HRR) gene function and homologous recombination deficiency (HRD) mediated by promoter methylation-driven biallelic loss of HRR genes in gynecologic cancer patients ( N = 27). For practical convenience, this method was also implemented on qPCR platform with high performance (0.5% limit of detection). With its adaptability and potential utility in ctDNA detection and treatment, this approach holds promise for advancing clinical diagnostics.

Accurate prediction of Homologous Recombination Deficiency (HRD) is vital for personalized cancer therapy, yet genomic assays are often costly and complex. Predicting HRD from Hematoxylin and Eosin (H&E) stained whole-slide images (WSIs) via deep learning offers an alternative, but model generalization across diverse cancer types remains challenging. We developed a three-stage transfer learning framework to improve HRD prediction across cancer types. After establishing baseline models on data from eight cancer types, the TCGA-BRCA model was selected as a pretrained model for transfer learning applied to eligible cohorts. Transfer learning enhanced HRD prediction performance specifically in cancers with histological similarity to the TCGA-BRCA source model. The AUC increased by + 4% (TCGA-LUAD), + 5% (CPTAC-LUAD), and + 1% (SARC) compared to baseline models. Interpretability analysis confirmed that the model’s predictions were driven by histologically relevant tumor regions, which was further supported by quantitative cellular analysis showing that HRD-high status was characterized by a significantly higher density of neoplastic cells (HRD-high: median 16.0, IQR 1.0–27.0 vs. HRD-low: median 1.0, IQR:0.0–5.0; p < 0.001), whereas HRD-low cases featured a stromal-rich microenvironment. Conclusions This study demonstrates that transfer learning can enhance HRD prediction from H&E images for cancer types sharing histopathological features with the TCGA-BRCA source model, offering a more efficient and accessible approach for clinical HRD assessment.