Research On Predicting Protein-protein Interactions Based On Relevance Vector Machine

Protein-protein interactions play an important role in many cellular processes.Studies on protein-protein interactions and protein function will facilitates the understanding of live activities,clinical therapeutics,and pharmaceutical design.As a result,it is much important for studying and developing effective computational methods used to predict protein-protein interactions.Focusing on two essential aspects of protein-protein interactions based on computational methods,this dissertation refers to the study of feature extraction method and classification algorithm.The main innovations are elaborated as following:(1)The research on feature extraction method based on protein sequences.This dissertation proposed a feature extraction method based on local protein sequence PSSM matrix coding and serial multi-feature Fusion.The method can capture proteinprotein interaction information of continuous and discontinuous for protein sequence by using the local coding;much key feature information contained protein sequences can be integrated through employing serial multi-feature Fusion.The comparison experiment was executed between the proposed method and others feature extraction methods on yeast and human dataset,the experimental results proved that the proposed method is robust and effectiveness.(2)The research on supervised prediction classifier based on protein-protein interactions.This dissertation proposed a classification algorithm of compound kernel function RVM based on gray wolf optimization algorithm and K-fold cross Validation.The algorithm can obtain the optimal solution of RVM kernel function width by using the intelligent optimization algorithm based on gray wolf optimization algorithm and K-fold cross Validation;the compound kernel function of RVM was created based on local Gaussian kernel and global quadratic polynomial kernel,which not only overcome the drawback of low prediction precision that was caused by the operation mode of RVM single kernel function,but also fully consider the special features of local and global of protein-protein interactions position.In the experiment,we not only compared the proposed intelligent optimization algorithm with others algorithms,but also the comparison was carried out between compound kernel function RVM and single kernel function RVM.In addition,we compared the prediction results between the prediction model created by the proposed method and other prediction models.The experimental results demonstrated that the proposed supervised classification algorithm is effectiveness.(3)The research on semi-supervised prediction classifier based on protein-protein interactions.This dissertation proposed a self-training semi-supervised classification algorithm of RVM based on AP clustering and Renyi entropy fusion.The algorithm can greatly reduce the influence of noise data on the prediction accuracy of classifier by using AP clustering and Renyi entropy fusion to assign labels for unlabeled samples;the semi-supervised classifier with optimal performance was constructed through adding the unlabeled samples with high degree of confidence to the training set and executing the self-training iteration classification with the expanded training set.It is demonstrated that the proposed semi-supervised classification algorithm is effectiveness by experimenting validation on M.musculus,H.pylori and H.sapiens datasets.(4)The online prediction system on protein-protein interactions was designed and developed,the system provide the supervised classification prediction on yeast and human datasets,and semi-supervised classification prediction on M.musculus,H.pylori and H.sapiens datasets,respectively.