The Prediction Method Study Of Protein-protein Interaction Based On Hybrid Kernel Function Of SVM

With human and other living species sequencing projects are completed, the biology and life sciences that humans study are going into the post-genomic era from genomic era and the human will shift the research focus proteomics that centered on protein interaction. The protein mainly achieves its function through interacting with other proteins and other Biological macromolecules, so the protein interaction will promote further research and the understanding of lives activities protein function. In recent years, with the rapid development of biological experiments, humans begin use a variety of experimental methods to study protein interaction. However, due to the constraint of current biological experiments and high-throughput experimental techniques and it often brings much high rates of both false positive and false negative predictions. Therefore, this paper explores the protein interaction from a computational point of view, and focuses on the prediction method of protein-protein interaction and protein interacting sites. The work of this paper mainly includes some aspects as following:In order to improve the accuracy of protein interacting sites’ prediction method, this paper using the sliding window technique combined with SVM base on hybrid kernel function of gaussian kernel and polynomial kernel and fully considering the impacts of adjacent or connected protein. At the same time this paper using the method of Booststrap to make a problem of unbalanced data protein interaction sites into balanced data problem, and the improved particle swarm optimization algorithm is applied into the SVM with hybrid kernel function and selective ensemble algorithm. It also improves the efficiency and the accuracy of prediction. Finally, testing in a real data set show that the new methods we put have a better effect and can improve the accuracy of protein interacting sites’ prediction method effectively.However, with the deepening research of biology, this paper puts forward a kind of SVM based on the pairwise hybrid kernel function of cosine kernel and linear differential accumulate kernel for protein-protein interaction prediction problem. This method considers the feature of protein domains fully. At the same time, according to the data of the protein-protein interaction should be have a feature of sequence-independent, so the idea of the "pairwise" take into the SVM kernel function. Testing on two real data of Yeast PPI and Human PPI, and the results show that the new method in this paper can improve the accuracy of predicting protein-protein interaction effectively compared with other methods.