A Statistical Framework for Evolutionary Analysis of Recurrent Somatic Mutations in Cancers

Current cancer genomics databases have accumulated millions of somatic mutations that remain to be further explored, faciltating enormous high throuput analyses to explore the underlying mechanisms that may contribute to malignant initiation or progression. In the context of over-dominant passenger mutations (unrelated to cancers), the challenge is to identify somatic mutations that are cancer-driving. Under the notion that carcinogenesis is a form of somatic-cell evolution, we developed a two-component mixture model that enables to accomplish the following analyses. (i) We formulated a quasi-likelihood approach to test whether the two-component model is significantly better than a single-component model, which can be used for new cancer gene predicting. (ii) We implemented an empirical Bayesian method to calculate the posterior probabilities of a site to be cancer-driving for all sites of a gene, which can be used for new driving site predicting. (iii) We developed a computational procedure to calculate the somatic selection intensity at driver sites and passenger sites, respectively, as well as site-specific profiles for all sites. Using these newly-developed methods, we comprehensively analyzed 294 known cancer genes based on The Cancer Genome Atlas (TCGA) database.