Using Entropy Cluster-Based Clustering for Finding Potential Protein Complexes

Many researches have studied the complex system today because protein complexes, formed by proteins that interact with each other to perform specific biological functions, play a significant role in the biological area. And a few years ago, E.C. Kenley and Y.R. Cho introduced an algorithms which uses the entropy of graph for clustering in [2,3] based on protein-protein interaction network. In our study, we extend the works to find potential protein complexes while overcoming existing weaknesses of their algorithms to make the results more reliable. We firstly clean the dataset, build a graph based on protein-protein interactions, then trying to determine locally optimal clusters by growing an initial cluster combined of two selected seeds while keeping cluster's entropy to be minimized. The cluster is formed when its entropy cannot be decreased anymore. Finally, overlapping clusters will be refined to improve their quality and compare to a curated protein complexes dataset. The result shows that the quality of clusters generated by our algorithm measured by the average cluster size considering f1-score is spectacular and the running time is better.