Reliable Protein-Protein Interaction Network Constructing And Its Application In Complexes Identification

In protein-protein interaction(PPI)networks,identifying protein complexes helps to reveal the function of protein and understand specific biological process.So it is of very important significance for the study of complex diseases.How to detect protein complexes that have biological significance from PPI data has become a breakthrough for uncovering various functions of organisms.The development of high-throughput experimental technologies provides us with a large number of PPI data,which makes it possible for us to apply computational method to identify protein complex in PPI networks.However,due to the limits of experimental environment and deviation,PPI data from high-throughput experimental technologies are often noisy,which affects the performance of the complexes identification algorithm.In order to decrease the high false positive and false negative rate in protein interaction data,this thesis use GO semantic similarity measure combining with protein experimental dataset to evaluate the reliability of the PPI edge,taking the biological functions of proteins into account.Then we evaluate the reliability of the protein-protein interaction to reconstruct the experimental protein interaction network.We apply several algorithms into this reconstructing reliable network,and the experimental results demonstrate that the reliable evaluation method can help these algorithms to improve their identification precision and the percent of identified complexes that have significant biological information are increased significantly.The thesis proposes a complexes identifying method in weighted network based on Gene Ontology and edge clustering coefficient(MWGE).Firstly,this method uses GO semantic similarity measure cooperating with edge clustering coefficient to evaluate the reliability of the PPI edge,generating weighted network.Then,we iteratively re-execute the simulating stochastic flows to identify complexes in the weighted network.Finally we use ?-module and overlapping-rate to remove low-quality and redundant complexes.The experimental results show that the complexes identified by MWGE method can match with the benchmark complexes better.Moreover,MWGE method outperforms other methods,especially in precision and comprehensive evaluation metric.