VPTMdb: a viral post-translational modification database

Yujia Xiang; Quan Zou; Lilin Zhao

However, the combination of EGAAC, BINARY, ZSCALE and CTriad features did not significantly enhance the model's performance, which suggests that high-dimensional features may include useless features that weaken the model performance. Among all the features, considering the three evaluation values of F1-score, MCC, AUC and dimensions, the AAC combined with the ZSCALE performed best, and the sensitivity, AUC and F1-score were higher than the single z-scale features. The independent test also shows that AAC combined with ZSCALE features significantly increased the AUC, F1-score, MCC, and Sn by 0.90%, 21.7%, 2.60%, and 25.0%, respectively (Supplementary Table S2 ). Now, it is important to answer two questions: (i) what is the difference between phosphorylated sites and non-phosphorylated sites and (ii) which features contribute most to the viral phosphorylation protein? To this end, we analysed the z-scale feature information between phosphorylated sites and non-phosphorylated sites. Then, we selected the most important features from the combined features with the mRMR method and using svm, random forest and naïve Bayes to perform a predictive evaluation.