Prediction Of Protein Function Based On Community Structure And Bay Esian Network

Proteins are the ultimate executer of cells’function. Protein is the key factor for any species to maintain life activities. Thus, the research of protein’s function, as a promising area for the research of proteome, can help people to have better understanding life. High-throughput experiment has already provided large amount of data related with proteins’function and protein-protein interaction. Those data, however, contains a relatively high false positive, thus it is very inefficient to testify the accuracy of those data by biological experiment. As a result, predicting proteins’function by using the data mining and clustering method from filed of computer science has become a hot topic in the field of bioinformatics.The protein-protein interaction data can build a network, thus we can use the method of clustering to analysis protein-protein interaction network. So far, most of the clustering method divide the network into several independent part, which ignores the fact there might be some overlapping part between different community within the network. The Clique Percolation Method proposed by Palla et al. explores the complete connected subgraph with the network, and considering the issue of overlapping part of different community, which is more suitable for the research of protein-protein interaction network. We will use the CPM to analysis protein-protein interaction network and predict unannotated proteins’function. First, we build a comprehensive protein database by combine protein-protein interaction data from DIP and BioGrid, and Proteins’function data from MIPS. Then, we explore the overlapping community structure in protein-protein interaction network by using CPM. Finally, we will use the model of Bayesian network to predict unannotated proteins’function.The protein database is developed under Java, with Windows7. The result of experiment in this paper, by predicting unannotated protein’s function of Arabidopsis thaliana, shows that the method proposed by this paper can yield good result.