Efficient exploration of multi-cancer networks by generalized covariance selection and interactive web content

Other conference contribution, 2015

Statistical network modeling techniques are increasingly important tools to analyze cancer genomics data. However, current tools for network construction and interpretation are not designed to work across multiple diagnoses and technical platforms, thus limiting their applicability to comprehensive pan-cancer datasets such as the Cancer Genome Atlas (TCGA). Here, we describe a novel strategy to construct and analyze integrative network models heterogeneous data from multiple cancers. First, we introduce a generalization of sparse inverse covariance selection (SICS) designed to integrate genetic, epigenetic and transcriptional data from multiple cancers into a comparative network. The algorithm is shown to be statistically robust, effective at detecting direct pathway links in data from The Cancer Genome Atlas (TCGA), and uses a new strategy involving non-informative priors to balance different cancers and data types. Second, we propose to rationalize the interpretation of the derived networks by a new and publicly accessible tool (cancerlandscapes.org), in which derived models are explored as interactive web content, linked to several pathway and pharmacological databases. To evaluate the performance of the method, we constructed a model of genetic, epigenetic and transcriptional data for eight TCGA cancers, using data from 3900 patients. The derived model rediscovered known mechanisms and contained interesting predictions. Possible applications include the prediction of regulatory relationships between genes in particular cancers, comparison of network modules in across multiple forms of cancer, and identification of drug targets in relation to network structure.