Abstract 5062: Identification of synthetic lethal gene interactions to discover novel cancer vulnerabilities and guide drug development

Abstract

Abstract Poor treatment responses in patients and drug resistance are main challenges in the clinic which can arise due an incomplete understanding of genetic dependencies within cancers and an ill defined patient population. Identifying synthetic lethal (SL) relationships between a gene pair, where loss of either of the participating genes has no effect on cancer cells, however simultaneous disruption of both genes leads to cancer cell death, can uncover major survival mechanisms and expose vulnerabilities within cancers. In spite of advances in CRISPR or RNAi based loss of function screening it is difficult to identify these gene interactions solely by experimentation due to the vast mutational landscape of cancers, inability of cell lines to capture clinically relevant mutations, and large number of possible SL gene combinations. Furthermore, large scale CRISPR/RNAi screens are prone to false positive results, leading to erroneous identification of gene interactions. We developed a machine learning model to identify SL genetic interactions by integrating pathway, ontology, genomic, and loss of function features on all gene pairs. This included features, which have been shown to be predictive of SL pairs such as, involvement in important cancer hallmark processes or signaling pathways, co-expression profile in patient cancers, and many more. The model was trained using SL pairs from published large scale experimental screens across several cancer types, ultimately enabling the prediction of cancer agnostic SL pairs. Non-SL pairs had to be computationally defined using post-perturbation gene expression profiles due to the lack of published data annotating non-SL pairs. The model showed significant predictive performance (AUROC = 0.94, AUPRC = 0.93, Specificity = 0.91), additionally, when tested on a gold standard, biologically verified set of SL gene pairs across several cancer types, the model showed good generalizability. We leveraged the model to identify key clinically relevant SL relationships and uncover complementary mechanisms tied to pro-survival processes. By identifying SL gene interacting pairs, our platform can help gain a better understanding of dominant mechanisms driving cancers and guide the discovery of novel therapies, biomarkers, and combination therapies early on during drug development. Citation Format: Mukti Parikh, Manuel García-Quismondo, Neel S. Madhukar, Coryandar Gilvary. Identification of synthetic lethal gene interactions to discover novel cancer vulnerabilities and guide drug development [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5062.