Protein Function Prediction Based On Protein Interaction Network And Clusterinq Alqorithms

Protein function prediction based on protein-protein interaction network is an important topic in the post-genomic era, in which a great deal of achievements has been acquired so far. One major branch of network-based computational methods for function annotation is to utilize clustering algorithms to divide the protein interaction network into function modules. However, the way of assigning protein functions according to function enrichment analysis directly on the basis of clustering modules cannot produce good results. It requires further researches so as to employ clustering algorithms appropriately. By means of combining clustering algorithms with other function predicting algorithms and integrating other sources of biological data, with proper models constructed, the complex information can be effectively adopted to improve the accuracy for protein function prediction.In this paper, with integration of protein domain information, after a proper modification of a domain context similarity based-algorithm, we redefined the formula of domain similarity and proposed a new algorithm PDSim that is based on parameterized domain combination similarity. The comparison with some known classical algorithms demonstrates a better predicting performance of our algorithm. Besides, after reuse of proteins’own composition domains to expand the scope of domain contexts while computing the domain combination similarity between two proteins, the impact of proteins’own domains is hence enhanced, and in such a way the modified RPDSim algorithm can work out even better results.Afterwards, we did some research and attempts on the application of clustering algorithms in predicting functions. Firstly, those clustering algorithm instances of MCL, MCODE, CFinder, DPClus and HC-PIN are taken as examples to annotate proteins in the light of generated complexes, which indicate the necessity of new forms of clustering algorithms involvement. Then, making use of complexes data produced by clustering algorithms or experimental methods, we extends the scope of domain context and thus proposed an algorithm DSCP that is based on clustering complex and domain information. A series of experimental tests on DSCP have manifested its accuracy as well as reliability in function prediction over simply complex-based predicting. As such, DSCP can be a feasible scheme for applying clustering algorithms to the prediction of protein functions.