FIG. 2.

The framework of PredHS. (A) Feature representation: We encode each interface residue using 108 site features, 108 Euclidean neighborhood features, and 108 Voronoi neighborhood features. (B) Two-step feature selection: the first step of feature selection is done by a random forest algorithm, we select the top 77 features with Z-Score larger than 2.5; the second step is performed using a wrapper-based feature selection. Features are evaluated by 10-fold cross-validation with the SVM (support vector machine) algorithm, redundant features are removed by sequential backward elimination. (C) Prediction models: PredHS-SVM and PredHS-Ensemble. For PredHS-Ensemble, an ensemble of n classifiers is built using different subsets, the final result is determined by majority votes among the outputs of the n classifiers.