Research On Prediction Of Protein-Protein Interactions Based Maximum Entropy Clustering Algorithm

All biological processes within cells invo1ve protein-protein interactions, meanwhile, the disorder of these interactions cause people sick, so analysing the interactions among proteins is not only important to uncover the mystery of the biological processes but also play an important role both in understanding the mechanisms of the diseases and the development of drugs.In the early research, people analyse protein-protein interactions mainly by experiments for the lack of data. However, these experiments are expensive and boring, people begin to analyse the protein-protein interactions by computation method as the increasing of the data, they analyse the feature of the interacting proteins and the noninteracting proteins, and then predict the interactions of proteins. Furthermore, people develop many algorithms, such as method based on random decision forest framework, method based on support vector machine, method based on Bayes and so on.The domain is the unit of structure and function of the proteins, they are a segment of conservative amino acid sequence and also an important structure and functional area for protein-protein interactions. Proteins interact with each other by special binding, and the binding site is precisely the domain, so, it is popular to think that the interactions among domains cause the interactions among proteins.In this paper, we analyse the feature of the interacting and noninteracting protein pairs, and present a method based on maximum entropy clustering algorithm to analyse and predict the protein-protein interactions. We test our method using the data in von Mering data set and the DIP database, the specificity is 94% and the sensitivity is 92%. Based the above method, we build a website (http://219.217.238.183:7001/prepi/index.jsp) to predict the interaction between proteins pairs.