Optimizing transcriptome-based synthetic lethality predictions to improve targeted therapy / immunotherapy treatment prioritization in non-metastatic breast cancer: BC-SELECT

Abstract

BackgroundPatients with non-metastatic triple-negative (TNBC) or HER2-positive (HER2+) breast cancer do not have tools to predict treatment response or prioritize options prior to starting therapy. Pathologic complete response (pCR) is a proxy that requires neoadjuvant treatment. BC-SELECT leverages genetic interactions (e.g. synthetic lethality (SL)/rescue (SR)) to predict treatment response via pCR from breast cancer gene expression data. BC-SELECT involves (1) "gene-pair identification" that identifies clinically relevant partner genes from large-scale datasets for a given targeted therapy or immunotherapy target, and (2) "training / parameter tuning" using clinical trials to predict treatment response. We evaluated BC-SELECTs ability to predict pCR for trastuzumab, poly (ADP-ribose) polymerase (PARP) inhibitors, and immunotherapy, with tuning supported by three trials (n=86) and validation on nine unseen trials (n=722). ResultsBC-SELECT significantly predicted pCR in 6/9 trials, including 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 significantly distinguished between pCR and residual disease in 8/9 trials. ConclusionsNo current standard-of-care, pre-treatment approaches predict benefit in non-metastatic TNBC and HER2+ breast cancer. BC-SELECT leverages genetic interactions to predict treatment response from gene expression data. Our findings support development of BC-SELECT towards prioritizing PARP inhibitors and immunotherapy in non-metastatic breast cancer.