Ranking novel cancer driving synthetic lethal gene pairs using TCGA data

Synthetic lethality (SL) has emerged as a promising approach to cancer therapy. In contrast to the costly and labour-intensive genome-wide siRNA or CRISPR-based human cell line screening approaches, computational approaches to prioritize potential synthetic lethality pairs for further experimental validation represent an attractive alternative. In this study, we propose an efficient and comprehensive in-silico pipeline to rank novel SL gene pairs by mining vast amounts of accumulated tumor high-throughput sequencing data in The Cancer Genome Atlas (TCGA), coupled with other protein interaction networks and cell line information. Our pipeline integrates three significant features, including mutation coverage in TCGA, driver mutation probability and the quantified cancer network information centrality, into a ranking model for SL gene pair identification, which is presented as the first learning-based method for SL identification. As a result, 107 potential SL gene pairs were obtained from the top 10 results covering 11 cancers. Functional analysis of these genes indicated that several promising pathways were identified, including the DNA repair related Fanconi Anemia pathway and HIF-1 signaling pathway. In addition, 4 SL pairs, mTOR-TP53, VEGFR2-TP53, EGFR-TP53, ATM-PRKCA, were validated using drug sensitivity information in the cancer cell line databases CCLE or NCI60. Interestingly, significant differences in the cell growth of mTOR siRNA or EGFR siRNA knock-down were detected between cancer cells with wild type TP53 and mutant TP53. Our study indicates that the pre-screening of potential SL gene pairs based on the large genomics data repertoire of tumor tissues and cancer cell lines could substantially expedite the identification of synthetic lethal gene pairs for cancer therapy.