Predicting Synthetic Lethal Genetic Interactions in Breast Cancer using Decision Tree

Abstract

Recently, a type of genetic interaction, termed synthetic lethality, is emerging as a potential promising anticancer strategy. Synthetic lethality indicates that simultaneous silencing of two complementary signaling pathways can cause cell death, while deficiency of any single gene will not show phenotype. In this study, we aimed to analyze and predict synthetic lethal genetic interactions based on decision tree in breast cancer using TCGA data. First, candidate gene pairs were collected using mutation data based on Misl algorithm, and involved genes were found in more than 2.5% total samples. Based on this method, we obtained 51,040 candidate gene pairs containing 320 genes. Second, 281 experimentally validated gene pairs were used to classify and optimize two features of mutation coverage and copy number variations (CNV) gain/loss, and the final integrated scores were used to predict synthetic lethal genetic interactions based on decision tree. Finally, candidate gene pairs were performed multi-level integrative analysis to search potential interactions, and 11,758 pairs were primarily identified. Some key gene pairs could be further screened based on drug responses and amplification features for experimentally identification, and we finally screened 5 gene pairs to perform further analysis. These results may contribute to screening and identifying synthetic lethal genetic interactions to uncover potential therapeutic target.