Clustering Analysis Of Multiple Protein-protein Interaction Data Based On Graph Theory

With the development of the science of life,proteomics has become an important research field.The analysis of protein-protein interaction network has become an important issue.Through the analysis of protein-protein interaction network,the protein complex and functional modules are excavated to reveal the law of life development.This issue mainly involves three parts: First,the formation of protein interaction data set;Second,methods of data analysis and mining;Third,evaluations of the data results.In this paper,we have collected protein-protein interaction data of 19 breast cancer genes with high mutation frequency.Protein-protein interaction data sets for high frequency mutations in breast cancer of all species(ABPPI)or human(HBPPI)is formed by integrating.Two data sets were clustered by K-medoids graph clustering algorithm and MCODE graph clustering algorithm.Two graph clustering algorithms have advantages and disadvantages.K-medoids graph clustering algorithm is sensitive to initialization,clustering results are diverse.MCODE graph clustering algorithm is not necessarily denseness.By using the Davies-Bouldin index parameters to evaluate the clustering results of the two algorithms,the DBindex average of MCODE algorithm is smaller,which shows its clustering effect is better;The DBindex variance of K-medoids algorithm is smaller,which shows that it is less affected by input parameters,and the clustering results are more stable.Based on the biological targets of proteins associated with breast cancer,the location of the protein in the resulting clustering is obtained to predict protein complexs.Next we will combine the advantages of the two algorithms,design a new graph clustering algorithm to update the central point in the high density region in order to obtain more effective clustering results.