Computational methods to dissect gene regulatory networks in cancer

Cancer is a disease of gene dysregulation, where cells acquire genetic alterations that drive aberrant signaling. These alterations adversely impact transcriptional programs and cause profound changes in gene expression. Large international consortia have generated massive tumor profiling data sets across many cancer types, collecting mutation and copy number variation, mRNA expression, and in some cases epigenomic and proteomic profiles. An overarching goal of these tumor-profiling efforts is to identify genes that are essential drivers of cellular processes in cancer. Here we review diverse computational methodologies that have sought to interpret somatic alterations and gene expression data through models of gene regulatory networks. Early work in the field used expression data alone to infer regulatory networks, and expression-only network inference continues to be an active area of research. Once catalogs of somatic mutations and copy number variations became available, another class of methods tried to interpret these alterations either in terms of a prior interaction network or an inferred regulatory network. More recently, the tools of regulatory genomics have been applied to cancer data sets, integrating regulatory sequence information and epigenomic data with gene expression, often through supervised methods. Finally, newer cross-cutting algorithms link upstream signaling changes - through a prior network or proteomics data - to downstream transcriptional programs and interpret somatic alterations in terms an integrative model. Currently, clinical trials use specific somatic alterations alone to direct patients towards pathway-targeted therapies. Ultimately, methods that interpret patient mutations profiles through the lens of gene regulatory networks may be better able to identify patient populations who will benefit from therapy.