VPTMdb: a viral post-translational modification database

Yujia Xiang; Quan Zou; Lilin Zhao

Five main procedures were performed to build the VPTMpre predictor. (i) a balanced benchmark dataset was constructed using the Synthetic Minority Oversampling Technique (SMOTE) (Chawla, et al., 2002) sampling method (Supplementary Material Si1); (ii) various feature representative methods were compared to obtain an effective feature representation strategy, with support vector machine used as the base classifier in a 5-fold cross-validation approach to find the best feature groups; (iii) the predictive performance of three classifiers (svm, random forest, naïve Bayes) on different feature dimensions was compared using the Minimum redundancy and maximum relevance (mRMR) method, and the features that performed well in all three classifiers were selected as the most meaningful and significant features; (iv) a 10-fold random independent test was performed to evaluate the predictive performance of the three different classifiers (svm, random forest, naïve Bayes); and (v) VPTMpre was implemented in the online web server.