Research On Protein Complexes Identification Algorithm Based On Intelligent Optimization Algorithm

With the implementation of the human genome project,biomedicine has entered the era of post-genomics,and it has become one of the research hotspots to systematically understand the laws of proteins that complete various life activities through interactions.As one of the complex networks,protein interaction networks have obvious community structures.These community structures usually correspond to specific functional modules called protein complexes.Identifying protein complexes from large-scale protein interaction networks plays an important role in predicting protein function and interpreting specific biological processes.The identification of protein complexes based on graph clustering algorithms is an effective method for identifying protein complexes by discovering community structures(also called clusters)in protein interaction networks.The graph clustering algorithm based on seed expansion strategy can effectively find most known protein complexes.However,on the one hand,the clustering results of such algorithms are greatly affected by the selection of seed nodes.On the other hand,once the expansion process is completed,The class result will no longer be adjusted.In this paper,we design a graph clustering algorithm from the perspective of intelligent optimization algorithm to find dense clusters in the protein interaction network,and then identify the protein complexes,mainly including the following two aspects:(1)A graph clustering algorithm based genetic algorithm(GAGC)was proposed.The representation of chromosome was designed.The IPCA seed node selection method was improved and the initial population was generated.The f-measure was selected as the objective function for the population evolution.The objective function is used to evaluate the chromosome quality;the chromosome alignment method is designed to perform crossover operations;and the clustering results are optimized through chromosome crossover and mutation.Compared with DPClus,MCODE,IPCA,ClusterOne,HC-PIN and CFinder,the experiments show that the algorithm can improve the diversity of the solution,and then expand the search space of graph clustering algorithm,and improve performance for the identification of protein complexes.(2)A graph clustering for protein complexes(DPSOPC)based on discrete particle swarm optimization is proposed.Firstly,particle position,velocity,objective function,and particle state update rules are defined.In the particle state update process,the particle adjust the cluster structure and optimize the clustering results.The global optimal solution of the particle swarm is used as the final protein complex recognition result.The experiments on five real Saccharomyces cerevisiae interaction networks were compared with DPSOPC,MCODE,IPCA,ClusterOne and other algorithms.The experimental results show that the algorithm can achieve high recall values.