Multicentre validation of BRCAura, an AI-based histology model predicting germline BRCA1/2 mutations in HR+/HER2− breast cancer

Germline BRCA1/2 alterations (gBRCAm) are actionable drivers of breast cancer management, yet testing strategies based on clinical and family-history criteria miss many carriers. We developed BRCAura, a deep learning model that predicts gBRCAm status directly from routine histology whole-slide images. The model was externally validated in a retrospective multicentre cohort of 1025 patients with early-stage hormone receptor-positive (HR+), Human epidermal growth factor receptor 2-negative (HER2−) breast cancer across six centres. BRCAura achieved area under the curve values of 0.81 for resections and 0.82 for biopsies, maintaining >93% sensitivity at a specificity of approximately 40%, consistent with its intended high-sensitivity rule-out design, while ruling out approximately one third of wild-type cases. Performance was robust across staining protocols and scanners and significantly exceeded established clinical and family-history models. Phenotype analysis revealed enrichment of immune infiltration and increased tumour cellularity in mutation-associated regions. These findings suggest BRCAura could help streamline germline testing pathways by optimising sequencing allocation and enabling efficient histology-based pre-screening.