| Post-translational modification(PTM)is essential for many biological processes.Potential post-translational modifications of proteins are composed of the central sites and adjacent amino acid residues.Covalent and generally enzymatic modification of proteins can impact the activity of proteins.Modified proteins would have more complex structures and functions.Post translational modifications of proteins contribute to biological functions and help us to understand the molecular mechanisms of protein design and drug design.Knowing whether a specific residue is modified or not is significant to unravel the function and structure of this protein.As experimental approaches to discover protein PTM sites are always costly and time consuming seriously inconsistent with the explosive growth rate of biological data,so computational prediction methods are desirable alternative methods.Nitrotyrosine is one of the post-translational modifications(PTMs)in proteins that occurs when their tyrosine residue is nitrated;Lysine phosphoglycerylation is a type of newly discovered PTM that is related to glycolytic process and glucose metabolism.This paper analysed how to improve the accuracy the two PTMs of using computational methods to predict PTM sites.A variety of predict methods using machine learning algorithms to predict protein nitrotyrosine and phosphoglycerylation.In this paper,we collected various experimental verificated nitrotyrosine protein sequences and phosphoglycerylation sequences to build our predictors.Those sequences are divided into positive and negative samples.According the characteristics of multiple protein sequences,we applied various sequence features to convert protein sequences to digital vectors.A new computational prediction method was designed by using the characteristics of multiple protein sequences.The SVM machine learning algorithm was used to establish the prediction model.To evaluate the performance of predictors,we used jackknife-cross validation and independent test set validation to analyze the accuracy of the prediction model.According to the lateral alignment of the accuracy for each feature prediction,we studied the influence of various characteristics of the prediction results,attempted to analyze the biological mechanism of different modification.The importance of different features was analyzed by F-score,and several important features were found,which helped to analyze the mechanism of protein modification sites.Experiments show that different modified sites need to extract features to establish a specific prediction model according to their own characteristics,which may achieve a better prediction effect. |
