Additional file 2: of Protein-protein interface hot spots prediction based on a hybrid feature selection strategy QiaoYanhua XiongYi GaoHongyun ZhuXiaolei ChenPeng 2018 Supplementary Information for Protein-protein interface hot spots prediction based on a hybrid feature selection strategy. This file provides all the features generated in this study, and other tables for analysis and discussion. Table S4. All 82 features generated in the study. Table S5. The numerical values of 10 different kinds of properties of the 20 amino acids. Table S6. Features selected from 82 features and the corresponding cross validation performance in SFS process. Table S7. The top 11 normalized features selected by decision tree, F-score and mRMR. Table S8. Features selected and the corresponding cross-validation performance in PSFS process for normalized features. Table S9. Consensus results based on combining any two of the five models (MINERVA2, APIS, KFC2a, KFC2b, Our model). Table S10. Interface information referred to the interfaces in the independent test set. Table S11. Statistical performance of our model for predicting hotspot of the independent test set by the types of protein-protein interfaces. Table S12. The top 10 features selected by decision tree, F-score and mRMR. Table S13. Features selected and the corresponding cross-validation performance in PSFS process for the 48 old features reported in Xia et al.’s paper. Figure S1. The ROC curves for cross-validation results of the training data set and the predictive results of the independent test set. Figure S2. The F-measures based on different number of normalized features selected by different methods. A. F-measures on the cross validation; B. F-measures on the independent test set. Figure S3. The F-measures based on different number of features of the 48 old features selected by different methods. A. F-measures on the cross validation; B. F-measures on the independent test set. (PDF 10501 kb)