VPTMdb: a viral post-translational modification database

Yujia Xiang; Quan Zou; Lilin Zhao

Moreover, PTM was predicted in other species with machine learning approaches (He, et al., 2018; Huang and Li, 2018) . For viral protein serine modification sites identification, we implemented a novel feature-based classifier named VPTMpre into the VPTMdb to provide users with the ability to find viral protein phosphorylated sites. First, we compared several feature extraction methods using support vector machine via a 5-fold cross-validation method to obtain the best feature representative strategy. Second, for feature selection, we separately input the features extracted from the previous step into three machine learning predictors (support vector machine, random forest and naïve Bayes). Using the minimum redundancy and maximum relevance (mRMR) algorithm (Hanchuan, et al., 2005) , the predictive performance of these three classifiers on different feature dimensions was compared. Subsequently, the features that performed well in all three classifiers were selected as the most meaningful and significant features and 68-dimensional features were obtained. The results of independent testing showed that 68D features performed author/funder. All rights reserved. No reuse allowed without permission.