Optimizing transcriptome-based synthetic lethality predictions to improve precision oncology in early-stage breast cancer: BC-SELECT

Patients with early-stage triple-negative (TNBC) or HER2-positive (HER2+) breast cancer do not have tools to predict response before treatment. Pathologic complete response (pCR) is a proxy that still requires neoadjuvant treatment. BC-SELECT is a clinically focused computational tool that leverages genetic interactions, such as synthetic lethality (SL) or rescue (SR), to predict treatment response from gene expression data. BC-SELECT builds on existing related tools (SELECT, ENLIGHT) to navigate competing treatment options.

BC-SELECT involves two main steps: (1) a “gene pair identification” step that identifies clinically relevant partner genes using in vivo and in vitro datasets for a drug target, and (2) “training/parameter tuning” using early-stage breast cancer clinical trials to develop predicted response scores to treatment. We evaluated BC-SELECT’s ability to predict responders to trastuzumab, poly (ADP-ribose) polymerase (PARP) inhibitors, and immunotherapy by training on three trials including 86 patients and validating on nine trials including 722 patients.

BC-SELECT significantly predicted treatment response in 6/9 early-stage trials, including in all PARP inhibitors (ROC-AUCs 0.6-0.8; Odds Ratio (OR): 1.9-3.5) and immunotherapy (ROC-AUCs 0.7-0.8; OR: 3.2-7.1). BC-SELECT distinguished between responders and non-responders in 8/9 trials (FDR-corrected Wilcoxon rank-sum test p-value <0.1). BC-SELECT outperformed SELECT and ENLIGHT in predicting immunotherapy response. Specificity was comparable across subtypes, with HER2+ 0.64, TNBC 0.66, and hormone-receptor positive (HR+) 0.76.

There are no standard-of-care pre-treatment predictors of response in TNBC and HER2+ breast cancer. Our findings position BC-SELECT as an eventual decision-support tool to prioritize treatments approved for patients with early-stage breast cancer.

STATEMENT OF SIGNIFICANCE

BC-SELECT is a clinically focused computational tool that leverages genetic interactions, such as synthetic lethality, to predict treatment response from gene expression data for PARP inhibitors and immunotherapy in early-stage breast cancer.