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

Abstract

ABSTRACTIntroductionPatients with early-stage breast cancer, especially in the triple-negative and HER2-positive subtypes, do not yet have approaches to predict relative benefit from therapies before treatment. SELECT1is a computational tool that leverages genetic interactions, such as synthetic lethality or rescue (SL or SR), to predict treatment response from tumor transcriptomics data. SELECT was tested on endocrine therapy and HER2-directed therapy, leaving open the application to treatment options used in triple-negative breast cancer, such as PARP inhibitors or immunotherapy. We adapt SELECT to address this clinical context, broadening the application of transcriptome-based genetic interactions in patients with early-stage breast cancer.MethodsSELECT involves two main steps: (1) a “training” step that, for a given drug target, constructs a library of clinically relevant candidate SL/SR partner genes, and (2) a “testing/validation” step that calculates patients’ predicted response scores in unseen studies based on the relative expression of the candidate partner genes. In our approach, BC-SELECT, we apply breast cancer-specific expression data from breast cancer trials and datasets to both the training and testing steps.ResultsBC-SELECT predicted responders and non-responders in 11 clinical trials in breast cancer not used for training, with predictions for 7 of 11 trials (including all PARP inhibitor and immunotherapy trials) demonstrating statistically significant odds ratios using multiplicity-corrected Fisher exact tests. BC-SELECT not only outperformed its score components (when each was compared to chance), but also outperformed SELECT.ConclusionIn the early-stage setting, particularly given that there are no pre-treatment predictors of response in triple-negative and HER2-positive breast cancer, BC-SELECT predicts patients anticipated to respond to PARP inhibitors and immunotherapy. These findings underscore the algorithm’s potential as a decision-support tool to prioritize treatments for patients with early-stage breast cancer.