Study On Protein Post-Translational Modification Prediction Based On Serine/Threonine Site-Modification Network

Post-translational modifications (PTMs) regulate many aspects of biological behaviours including protein-protein interactions and cellular processes. PTMs are related to many diseases such as cancers and diabetes. So the PTM site prediction is of great significance for understanding the life processes, the prevention and treatment of diseases.The PTMs on serine and threonine sites include phosphorylation, O-linked glycosylation and acetylation. With the development of experimental techniques, a large number of experimentally verified PTM data have been accumulated, since the experimental approaches are high-cost and labor-intensive, which promotes the rapid development of computational methods. Most of computational methods only employ protein local sequence information and focus on a single PTM. Researches show that in situ PTM can reflect the potential functional relations between different PTMs. By employing this information in the prediction of the PTM sites, the performance of the prediction can be improved largely. In this work, we propose a site-modification network (SMNet) which can reflect in situ PTM information. By employing the SMNet profiles, the PTM sites on serine and threonine are predicted in this study. The details are as follows:Firstly, the human PTM data are obtained from various PTM databases, from which the in situ PTM information is collected, the SMNet is built by incorporating in situ PTM information. Then, by employing SMNet profiles as well as local sequence information, the computational approach named SMNet-SVM method trains the model and predicts the PTM sites based on the Support Vector Mechine (SVM). The SMNet-SVM method can predict different types of PTMs, and the 10-fold cross-validation is adopted to evaluate the prediction performance in this work. The result analysis suggests that the SMNet profiles play an important role in predicting the PTM sites and have improved the prediction performance significently. What is more, the SMNet-SVM method is also compared with other existing PTM prediction approaches. The results suggest that the proposed method performs better than other methods. Lastly, the potential PTM sites predicted by the SMNet-SVM method are further verified by documents, which suggests that our method is effective and can discover the potential PTM sites.Moreover, to further improve the prediction performance, including the SMNet information, the gene ontology from AmiGO and the protein-protein interaction from STRING are further employed in this study. Then by incorporating these features, the prediction of the PTM sites is performed based on multiple kernel learning and kernel ridge, and the results suggest that the protein functional information can further improve the performance of PTM site prediction.