Sequence-Based Prediction of Hot Spots in Protein-RNA Complexes Using an Ensemble Approach

RNA-binding hot spots are dominant and fundamental residues that contribute most to the binding free energy of protein-RNA interfaces. As experimental methods for identifying hot spots are expensive and time-consuming, high-efficiency computational approaches are required in predicting hot spots on a large scale. In this work, we proposed a sequence-based machine learning method to predict hot spots in protein-RNA complexes. We extracted 83 relative independent physicochemical features from a set of the 544 properties in AAindex1. Each physicochemical feature was combined with the predicted relative accessible surface area (RASA) and substitution probability feature from Blocks Substitution Matrix (BLOSUM) for training models by support vector machine (SVM) and k-nearest neighbor algorithm (k-NN). The combinations of the 166 individual models were explored and 33 top-performance models were selected to construct the final ensemble classifier by a majority voting technique. The ensemble classifier outperformed the state-of-the-art computational methods, yielding F1 score of 0.742 and AUC of 0.824 on the independent test set.