Abstract
Cancer is a complex disease including a range of disorders that are activated simultaneously by multiple biological processes on multiple levels. Various genome-wide profiling techniques have been developed to capture the dynamics of these processes at the epigenomic, transcriptomic, and proteomic levels. Integrative analysis of data from these sources has the potential to differentiate cancer subtypes from a holistic perspective that reveals connections that otherwise cannot be detected using observations from a single data type. In this article, we present a novel approach named DSCC (Disease Subtyping using Community detection from Consensus networks) that is able to discover disease subtypes from multi-omics data and is robust against noise. In an extensive analysis using simulation studies and 5,782 real patients belonging to 20 cancer datasets from The Cancer Genome Atlas, we demonstrate that DSCC outperforms state-of- the-art methods by correctly identifying known patient groups and novel subtypes with significantly different survival profiles.