A Combination Method of Centrality Measures and Biological Properties to Improve Detection of Protein Complexes in Weighted PPI Networks

Introduction: In protein-protein interaction networks (PPINs), a complex is a group of proteins that allows a biological process to take place. The correct identification of complexes can help better understanding of the function of cells used for therapeutic purposes, such as drug discoveries. One of the common methods for identifying complexes in the PPINs is clustering, but this study aimed to identify a new method for more accurate identification of complexes. Method: In this study, Yeast and Human PPINs were investigated. The Yeast datasets, called DIP, MIPS, and Krogan, contain 4930 nodes and 17201 interactions, 4564 nodes and 15175 interactions, and 2675 nodes and 7084 interactions, respectively. The Human dataset contains 37437 interactions. The proposed and well-known methods have been implemented on datasets to identify protein complexes. Predicted complexes were compared with the CYC2008 and CORUM benchmark datasets. The evaluation criteria showed that the proposed method predicts PPINs with higher efficiency. Results: In this study, a new method of the core-attachment methods was used to detect protein complexes enjoying high efficiency in the detection. The more precise the detection method is, the more correct we can identify the proteins involved in biological process. According to the evaluation criteria, the proposed method showed a significant improvement in the detection method compared to the other methods. Conclusion: According to the results, the proposed method can identify a sufficient number of protein complexes, among the highest biological significance in functional cooperation with proteins.