HRDPath: an explainable multi-model deep learning architecture for predicting homologous recombination deficiency from histopathology images.

Published in Nature communications. Abstract: Homologous recombination deficiency (HRD) is a critical biomarker for guiding treatment decisions in high-grade serous tubo-ovarian carcinoma (HGSOC), a cancer with few reliable biomarkers. However, existing genomics-based tests for HRD are variable, expensive, and time-consuming. We develop HRDPath, a patient-level deep learning architecture that predicts HRD status from whole slide images, requiring no additional molecular testing. HRDPath is validated across three cohorts of female patients and achieves an AUC of 0.846, significantly outperforming previously reported results in HGSOC. By incorporating multiple slides per patient and targeted optimization strategies, we strengthen the model clinical utility, with both specificity and sensitivity reaching 0.90. Our interpretability pipeline enhances model trustworthiness and reveals important clinical and biological insights into HRD-positive cancers, highlighting the associated morphological and pathological changes at the cellular and tissue levels. HRDPath provides a foundation for an accessible and scalable digital biomarker to improve treatment stratification in ovarian cancer globally.