Abstract
Precision medicine for cancer involves identifying and targeting the somatic genome alterations (SGAs) that drive the development of an individual tumor. Much of current efforts at finding driver SGAs have involved identifying the genes that are mutated more frequently than expected among a collection of tumors. When these population-derived driver genes are altered (perhaps in particular ways) in a given tumor, they are posited as driver genes for that tumor. In this technical report, we introduce an alternative approach for identifying causative SGAs, also known as “drivers”, by inferring causal relationships between SGAs and molecular phenotypes at the individual tumor level. Our tumor-specific causal inference (TCI) algorithm uses a Bayesian method to identify the SGAs in a given tumor that have a high probability of regulating transcriptomic changes observed in that specific tumor. Thus, the method is focused on identifying the tumor specific SGAs that are causing expression changes that are specific to the tumor. Those SGAs that have a high probability of regulating transcriptomic changes related to oncogenic processes are then designated to be the putative drivers of the tumor. In this paper, we describe in detail the TCI algorithm and its implementation.