Clustering of protein-protein interactions (PPI) and gene ontology molecular function using Markov clustering and fuzzy K partite algorithm

Abstract

Functional disorders of proteins in the human body can cause a certain disease. The function and role of the protein are represented by Gene Ontology (GO). In this study, the GO molecular function was used to enrich the analysis of protein-protein interaction (PPI). The relationship between PPI and GO molecular function was represented in a bipartite graph. In the pre-processing step, the PPI network was reduced using the Markov clustering algorithm to obtain the group of proteins with the highest modularity score. The fuzzy k partite algorithm was used to cluster the PPI network and GO molecular functions into several groups. The result of Markov clustering showed the accuracy of 84.6% compared to that of the same algorithm using the GIANT package on Cytoscape applications, one of the popular software for network analysis. Proteins obtained from Markov clustering results were used as inputs to obtain their related GO molecular function. Their relationship was represented as a bipartite graph which is used as an input for the fuzzy k partite algorithm. With the dataset of Diabetes Mellitus type II, the results of Markov clustering showed that there were 117 proteins and 328 related GO molecular function. With fuzzy k partite algorithm, the minimum cost for the bipartite graph is 594.175 at the 20 clusters of proteins and 29 clusters of GO molecular functions.