Identifying similar functional modules by a new hybrid spectral clustering method.

Recently, a large number of researches have focused on finding cellular modules within protein-protein interaction networks. Until now, most of the works have concentrated on finding small modules and protein complexes. The authors have extended the concept of functional module and have identified larger functional modules which are the most similar to the entire network. To this end, a new hybrid spectral-based method is proposed here. First, the original graph is transformed into a line graph. Next, the nodes of the new graph are represented in the Euclidean space by using spectral methods and finally, a self-organising map is applied to the points in the new feature space. The experimental results show that similar modules, obtained from the proposed method, have own local hubs and lots of significant functional subunits concerning each other. These modules not only detect general biological processes that each protein is involved in, but also due to great similarities to the original network, it can be used as significant subnetworks for predicting protein function as detailed as possible. Some interesting properties of these modules are also investigated in this research.