SiS: Significant subnetworks in massive number of network topologies

Availability of abundant biological network data and its noisy nature necessitates extracting reliable subnetworks hidden in it. One key step towards that goal is to discover the subnetworks that appear frequently in a collection of networks. This paper presents a method, named SiS (Significant Subnetworks), to discover most probable subnetworks (i.e., subnetworks that have the highest chance to exist) in a large collection of biological networks where each node is labeled with the corresponding molecule (such as gene or protein). SiS builds a template network which summarizes the entire set of input networks. It then grows subnetworks that are most probable with the guidance of this template network. Our experiments demonstrate that our method scales to very large datasets and subnetworks easily. On the metabolic networks of the eukaryote organisms, our method runs from a few seconds to a few minutes depending on the subnetwork size. MULE, an existing method for the same problem, takes hours or does not complete for days on the same dataset. Our results also suggest that the most probable subnetworks are often the most frequent ones as well.