Study On Post-Translational Modification Site Prediction Based On Protein Phosphorylation Related Site-Modification Network

Protein post-translational modification(PTM)is a very important way of regulating life activities,which can change the protein structure and improve the protein function.Therefore in-depth study of the principle,type and action mechanism of protein PTMs is important for understanding pathogenesis of human diseases.Recently,with the continuous development of experimental technologies,a large amount of protein PTM site data are accumulated,which greatly promote the research development of protein PTMs.However,experimental methods are often time-consuming,labor-intensive and high-cost,thus there is a necessary need to develop efficient and accurate computational approaches for predicting PTM site and provide useful reference information for the following experimental work.Most of computational approaches use protein amino acid sequence information to make prediction,while ignore functional association information between different protein PTMs.Researches have indicated that in situ PTM denotes as multiple PTM types take place at the same position on the amino acid sequence of the same protein,and can reflect functional associations between different PTMs.Therefore,the multiple sites-multiple PTMs relationships of in situ PTM inspire us to largely make use of network topological structure information from the point of phosphorylation related site-modification network,and further apply it in PTM site prediction.The main work is summarized as follows:1.Using the PTM data on S/T/Y sites collected from several PTM databases to construct the phosphorylation related site-modification network.On the basis of this network,a network link prediction method based on resource allocation named SMNBI(site-modification network based inference)is proposed.SMNBI mainly uses the known link information in the network to predict the unknown relationships between sites and PTMs.The SMNBI is compared with other existing methods including network-based link prediction methods and phosphorylation site prediction related methods,and the results show that the phosphorylation related site-modification network plays an important role in predicting phosphorylation sites and significantly improves the predictive accuracy.2.In order to solve the problem of predicting potential PTMs for an isolated site node in the phosphorylation related site-modification network,MK-SVM(multiple kernels support vector machine)is also proposed for comprehensively predicting different PTMs on S/T/Y sites in this study.Firstly,a Gaussian interaction profile kernel and protein local sequence kernel are developed by using Gaussian kernel and amino acid substitution matrix BLOSUM62,respectively.Then multiple kernels are input into SVM to train and predict by means of linear weighted combination.MK-SVM effectively solves the problem of prediction of an isolated node in the phosphorylation related site-modification network and can comprehensively predict different PTMs on S/T/Y sites.Compared with several commonly used PTM prediction methods,the results suggest that MK-SVM achieves a good predictive performance for different PTMs including phosphorylation,O-GlcNAc,nitration and sulfation etc.