Abstract 4958: Optimizing transcriptome-based synthetic lethality predictions to improve precision oncology in breast cancer: BC-SELECT

Abstract

Abstract Introduction: Precision oncology currently focuses on targeting specific mutations in cancer driver genes as predictors of treatment response. Clinical application of tumor transcriptomics is similarly limited to using specific RNA-based fusions for using therapy. SELECT (Lee et al, Cell 2021) is a computational tool that characterizes synthetic-lethal and/or synthetic-rescue genetic interactions of potential drug targets by using expression data, then uses these interactions to predict drug response in cancer patients. SELECT was initially developed using pan-cancer data from The Cancer Genome Atlas (TCGA), with limited evaluation on agents in breast cancer. We therefore sought to optimize SELECT for patients with breast cancer in an adaptation named BC-SELECT. Methods: The standard SELECT comprises two steps: (1) Using an in-vitro screen, a patient tumor data screen, and a phylogenetic screen to build a library of clinically relevant candidate synthetic-lethal/synthetic-rescue gene partners; and (2) Generating scores based on gene partners’ expression levels in a given sample. Scores are then used to predict the likelihood of response to a given therapy. For BC-SELECT, we modified this framework as follows: We incorporated the METABRIC dataset, comprising 1,989 samples and used only the breast cancer-specific TCGA dataset (1,075 samples) (which outperformed training on pan-cancer TCGA). To enhance the survival screening process, we include up to 5 clinical covariates for each training dataset, and prioritized gene pairs based on survival. Furthermore, we introduced new scoring controls to confirm prediction accuracy. All parameters have been standardized to optimize performance. Results: BC-SELECT predicted treatment response in our initial cohort of breast cancer trials with immunotherapy or targeted therapy and expression data, where the same treatment/treatment target was present in more than one trial. BC-SELECT prediction was more accurate for immunotherapy and PARP inhibitors compared to trastuzumab. The AUC ranges for immunotherapy are between 0.67 and 0.83. For PARP inhibitors, the AUC ranges from 0.5 to 0.78, and for trastuzumab, it falls between 0.5 and 0.75. Discussion: BC-SELECT advances transcriptome-based synthetic lethality prediction for translational use in patients with breast cancer. This tool demonstrates analytic and clinical validity to begin to improve therapy stratification and prioritization in patients. Citation Format: Yewon Kim, Matthew Nagy, Rebecca Pollard, Nishanth Ulhas Nair, Lipika Ray, Eytan Ruppin, Padma Sheila Rajagopal. Optimizing transcriptome-based synthetic lethality predictions to improve precision oncology in breast cancer: BC-SELECT [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4958.