Affiliations: [a] School of Information Science and Engineering, Hunan City University, Yiyang, China
| [b] College of Communication and Electronic Engineering, Hunan City University, Yiyang, China
| [c] College of Computer Science and Electronic Engineering, Hunan University, Changsha, China
Correspondence:
[*]
Corresponding author. Buwen Cao, School of Information Science and Engineering, Hunan City University, Yiyang 413000, China. E-mail: [email protected].
Abstract: The identification of overlapping protein complexes in proteinprotein interaction (PPI) networks may elucidate cellular functional organizations and their underlying cellular mechanisms. Recently, many protein complex mining algorithms have been developed for PPI networks. However, the majority of available algorithms primarily depend on mining dense subgraphs as protein complexes, thereby failing to consider the inherent biological meanings between protein pairs. Thus, methods for identifying protein complexes using the biological significance hidden in edges need to be investigated. In this paper, we propose IK-medoids, an improved method that detects overlapping protein complexes from weighted PPI networks based on the rough fuzzy relationships between protein pairs. The presented algorithm is primarily based on the fuzzy relationship that obtains the non-overlapping protein substructure, and then K-medoids is executed from the proteins in the PPI network. Next, the similarity between one protein and each candidate complex is calculated to determine whether the protein belongs to one or multiple complexes with the ration of each similarity to maximum similarity. In the end, overlapped protein complexes are merged to form the final protein complexes. We apply the method to three PPI networks and validate the results using two reference protein complexes retrieved from public databases. Experimental results show that our method outperforms classical algorithms, such as ClusterONE, CMC, MCL, OSLOM, and RFC, and achieves ideal overall performance in terms of F-measure, sensitivity, and accuracy.
Keywords: PPI, protein complex, overlapping, K-medoids, Fuzzy relation
