Computational identification of RNA-Seq based miRNA-mediated prognostic modules in cancer.

Systematic identification of miRNA prognostic signature can help decipher the effects of biomarkers in cancer treatment. A number of previous studies have only characterized a single miRNA as a promising prognostic biomarker. There is currently a trend towards combining several miRNAs as a panel of prognostic signatures, but few attempt to explain the mechanism of miRNA combination. Throughout this paper we refer to “miRNA-mediated prognostic modules” and propose a novel computational approach called ProModule to analyze prognostic biomarkers from the module perspective. ProModule works in two main stages: it first uses univariate and multivariable Cox proportional hazards regressions to find individual miRNA biomarkers and then employs a clustering method to systematically detect miRNA-mediated modules with statistical prognostic significance. We applied ProModule to three data sets in bladder cancer (BLCA), breast cancer (BRCA) and liver cancer (LIHC), and identified several miRNA prognostic modules for each data set. We found that miRNA prognostic modules have more powerful prognostic value than individuals, while presenting coherent miRNA-miRNA expression as well as significant functional enrichment, and thus are likely to be biologically meaningful. Availability: ProModule is implemented in R and available at https://github.com/chupan1218/ProModule.