Identifying Protein Complexes And Predicting Disease Genes Based On Protein Domain

The proteins and the interaction networks formed by proteins are of great help to understand kinds of life activities. The complex protein-protein interaction network contains lots of valuable information, which need effective methods to analysis and mining. Meanwhile, recently the protein domains play an important role in the structure of protein, and provide indispensable information with biological significance. Therefore, this paper proposed new algorithms based on intelligence optimization to identify the protein complexes(modules), and combined the information about domains to predict the disease genes.Considering the limitation and preference of protein-protein interaction(PPI) network, this paper combined the information about domain and the topology to construct the PPI network. And compared it with the other two different ways of constructing the PPI network to testify the effectiveness of adding the information about domain. In the PPI network, integrated into the main idea of swarm intelligence optimization, a novel protein complexes(modules) identifying algorithm based on brain storming strategy was proposed, named algorithm IPC-BSS (Identifying Protein Complexes based on Brain Storm Strategy). Imitating the discussion process of human brain storming, the IPC-BSS algorithm designed two update ways to let the protein can move between different protein complexes or combined two close protein complexes, which resulted in obtaining better and better protein complexes. It effectively overcomes the drawback once the proteins which had been divided into a module have no chance to shuffle between modules that had been established in previous steps. Compared with other classic algorithms, the experimental results indicate that the IPC-BSS algorithm can efficiently identify protein complexes with more actual biological significance, and accurately find some protein complexes with relatively large size.The connections between domains will constuct the domain-domain interaction network. The network explains the interaction between proteins from spatial structure, which has closely relationship with PPI network and provides very important biological information. Thus, the domain-domain interaction network was seen as the bridge to link the PPI network and the disease phenotype interaction network, and utilized the model of random walk in these three networks, this paper put forward a new algorithm UThrRW(Unbalanced Three Random Walk) to predict the disease genes. The UThrRW algorithm not only can mine the topology information in each network, but also utilize the hiding information between networks. Theoretically the domain-domain interaction network enhances the connection between PPI network and disease phenotype interaction network, which results in improving the effectiveness of the algorithm. The experiment results of UThrRW are better than RWR, BiRW bl and UBiRW.