Spectral clustering for detecting protein complexes in PPI networks

Abstract

PPI(Protein-protein interaction) networks decomposition is of great importance for understanding and detecting functional complexes in PPI networks. In this paper, we study spectral clustering for detecting protein complexes, focusing on two open issues in spectral clustering: (i) constructing similarity graphs; (ii) determining the number of clusters. Firstly, we study four similarity graphs to construct graph Laplacian matrices. Then we propose a method to determine the number of clusters based on the properties of PPI networks. A large number of experimental results on DIP and MIPS PPI networks indicate that every similarity graph shows its strengths and disadvantages, and our finding of the number of clusters improves the cluster quality. Finally, compared with several typical algorithms, spectral clustering for detecting protein complexes obtains comparable performance.