MiMod: A New Algorithm for Mining Biological Network Modules

It is fundamental to discover frequent network modules from multiple networks as they serve gene co-expression modules, biological pathways, or protein complexes. The rapid accumulation of biological network data has enabled the discovery of frequent network modules. There have been a number of algorithms developed for the discovery of frequent network modules in the past years, but most of them do not perform consistently well evaluated by accuracy and efficiency. Therefore, it is essential to develop new algorithms capable of effectively and efficiently extracting the significant modules. We introduce MiMod, a robust algorithm for mining the frequent network modules based on enhancing the sensitivity to mine frequent network modules via introducing a so-called compatible graph with edge weight reflecting frequent degree of local subnetworks. Tested on 43 gene co-expression networks from GEO and 13 tissue-specific protein interaction networks from SNAP, the MiMod, respectively, discovered 4805 and 485 biologically important and frequent modules that are missed by others. In addition, we found that the likelihood for a network module to be biologically meaningful increases with its density and frequency. Moreover, we also found highly cooperative relationships in the modules of protein complex networks, thus demonstrating the necessity of revealing frequent modules from multiple networks.