Multi-protein Complex Detection by Integrating Network Topological Features and Biological Process Information

Predicting multi-protein complexes is becoming a central problem in system biology as it could be a way to reveal the biochemical functions of a protein. Most of the recently developed methods focus on topological information to detect multi-protein complexes. In this paper, biological and topological features characterizing protein complexes are extracted and used in conjunction with muti-class support vector machines to detect multiprotein complexes from the protein-protein interaction network. The proposed method was able to detect 76 complexes out of 81 reference complexes with high precision. In comparison with state-of-the-art methods, the evaluation results indicate that the applied method has great potential in detecting multi-protein complexes.