Research On The Classification Algorithms Of Protein-Protein Interaction

The prediction of protein interaction plays a very important role in all biological processes. Protein complexes are formed by interaction between proteins, and different biological processes are performed at the same time, including Enzyme catalysis, immune response, endocrine function and DNA replication. However, these methods are often time-consuming and expensive. In recent years, more and more high-through experimental methods have been developed to use to predicted the interaction of proteins. Especially, over the last decade, with the increasing number of these experimental methods, the computational methods are becoming more and more significant in predicting protein interaction.Protein-protein interaction provides important theoretical basis for the study of the mechanism of major diseases, disease treatment, disease prevention and development of new drug. In the critical diseases use which probe in the use of the protein interaction, the disease of the tumor is one of the most important diseases which threaten human health currently. In order to reduce the mortality of tumor, the early diagnosis and effective treatment is of great importance. But if tumor diagnosis only relies on the subjective recognition of doctors, it is often misdiagnosed. So the classification of tumor based on gene expression data has already become a focus of current research.Two novel methods are proposed in the dissertation. The one is the prediction of protein-protein interactions based on the meta sample and sparse representation, and another one is the classification of tumors based on probabilistic classification vector machine. In the study of protein-protein interaction classification, this literature places emphasis on the extraction of a meta sample which could reflect the intrinsic structure of the protein. In the study of classification of tumors, the key point is to extract the best subset of the tumor by using the combination of the probability classification vector machine and the DX feature selection method to predict the tumor types.The major research effort of this paper is:1.First of all, classify and summarize the existing calculation methods for protein interaction and tumor prediction, and describe the theory of different methods briefly.2.In this paper, we proposed a prediction method for protein-protein interaction based on the meta sample(MSRC)first. The original sample used in the traditional method is not of representativeness, which go against to the improvement of classification accuracy. So we can dig out the deep biological significance of the data by extracting the representative samples, namely metasample, which could capture the special structure inherent in the data. In this literature, we use the method of singular value decomposition (SVD) to reduce the dimension of protein interaction training data set and get the metasample. Then the new test samples are expressed as the linear combination of metasample. Finally, the sparse representation classification (SRC) algorithm is used to classify the protein-protein interaction data sets. Experiments are carried out on the extracted data sets, and compared with some traditional classification algorithm. The results show that the proposed method can improve the accuracy of prediction.3. Based on DX scoring feature selection method, a classification method of tumor gene expression profile data is proposed, which is based on the probability vector classifier (PCVM). This method classifies the leukemia and prostate tumor data sets first, and at the same time, the gene expression profile data were sorted. The DX feature selection algorithm was used to select a set of data as the training data set with a better classification performance, then the PCVM algorithm is used to classify the test set. In order to test the predictive effect of the method, we compared the method with some preferable classification performance algorithms, and compared to the results after combined with the DX feature selection method again. At the same time, we also compared the method with some other excellent feature selection methods. The comparative analysis shows that the method is effective.