Protein Complex Identification And Disease Genes Prediction Based On PPI Network

Identifying the modular structure of biological networks and forecasting genes that cause human genetic diseases have important research value in the field of bioinformatics. This paper devises a new algorithm identifying protein complex based on topological properties of protein-protein interaction network(PPI network), which proposes that the ownership of a protein can be defined by the closeness of its neighbors. And then, applying the new algorithm on human protein interaction network to predict the candidate genes that may cause diversity diseases.It contains two aspects specifically:First, PPI network is a kind of complex network. Inspired by the organization law of social network and the essential gene identification, this paper proposed a new protein complex mining algorithm NC-TDPINs(Neighbor Closeness base on Transient Dynamic Protein Interaction Networks) based on neighbor connecting closeness conception. The method NC-TDPINs first selects the node with larger cluster coefficient and its neighbors as the initial complex kernel, and then expands the initial kernel to get the final complex based on neighbor closeness strategy which indicates that the ownership of one given node will be defined by the closeness of neighbors distributing in different subgraphs. Compared with other state-of-art algorithms, NC-TDPINs can find more complexes with biological significance and obtain better accuracy.Second, it is known that complex diseases involve genetic mutations, regulation disorders. This paper proceeds from the angle of disease-gene set relationship, first builds module-module interaction network based on human protein interaction network, then adopts Mpagerank(Modules PageRank) algorithm to score these modules, and last statistics and permutates the candidate genes. Genes with similar function will work together in format of module that results in disease phenotypes. According to "guilt-by-association" principle, it can predict the candidate genes by their relation with the known genes in same modules. The experimental results indicate that Mpagerank algorithm outperforms NetScore, NetZcore, and fFlow algorithms on predicting disease genes.