DOMINO: a novel network-based module detection algorithm with reduced rate of false calls

Network-based module discovery (NBMD) methods have taken a central role in integrative analyses of omics data in modern bioinformatics. NBMD algorithms receive a gene network and nodes9 activity scores as input and report sub-networks (modules) that are putatively biologically meaningful in the context of the activity data. Although NBMD methods exist for almost two decades, only a handful of studies attempted to compare the biological signals captured by different methods. Here, we first set to systematically evaluate six popular NBMD methods on gene expression (GE) data and Gene-Wide-Association Studies (GWAS). Notably, testing Gene Ontology (GO) enrichment of modules obtained by these methods, we observed that GO terms enriched on modules detected on the real data were often also enriched after randomly permuting the input data. To tackle this bias, we designed the EMpirical Pipeline (EMP), a method that infers the empirical significance of GO enrichment scores of an NBMD solution by computing, for each term, a background distribution of scores on permuted data. We used the EMP to fashion five novel performance evaluation criteria for NBMD methods. Last, we developed DOMINO (Discovery of Modules In Networks using Omics) - a novel NBMD algorithm. In extensive testing on gene expression and genome-wide association study data it outperformed the other six algorithms. As it produces solutions with only a few non-specific GO terms, DOMINO can be used without empirical validation. EMP and DOMINO are available at https://github.com/Shamir-Lab/.