Detecting homologous recombination deficiency for breast cancer through integrative analysis of genomic data.

Abstract

Homologous recombination deficiency (HRD) leads to genomic instability, and patients with HRD can benefit from HRD-targeting therapies. Previous studies have primarily focused on identifying HRD biomarkers using data from a single technology. Here we integrated features from different genomic data types, including total copy number (CN), allele-specific copy number (ASCN) and single nucleotide variants (SNV). Using a semi-supervised method, we developed HRD classifiers from 1404 breast tumours across two datasets based on their BRCA1/2 status, demonstrating improved HRD identification when aggregating different data types. Notably, HRD-positive tumours in ER-negative disease showed improved survival post-adjuvant chemotherapy, while HRD status strongly correlated with neoadjuvant treatment response. Furthermore, our analysis of cell lines highlighted a sensitivity to PARP inhibitors, particularly rucaparib, among predicted HRD-positive lines. Exploring somatic mutations outside BRCA1/2, we confirmed variants in several genes associated with HRD. Our method for HRD classification can adapt to different data types or resolutions and can be used in various scenarios to help refine patient selection for HRD-targeting therapies that might lead to better clinical outcomes.