Boosting Prediction Performance of Protein–Protein Interaction Hot Spots by Using Structural Neighborhood Properties

Abstract

Binding of one protein to another in a highly specific manner to form stable complexes is critical in most biological processes, yet the mechanisms involved in the interaction of proteins are not fully clear. The identification of hot spots, a small subset of binding interfaces that account for the majority of binding free energy, is becoming increasingly important in understanding the principles of protein interactions. Despite experiments like alanine scanning mutagenesis and a variety of computational methods that have been applied to this problem, comparative studies suggest that the development of accurate and reliable solutions is still in its infant stage. We developed PredHS (Prediction of Hot Spots), a computational method that can effectively identify hot spots on protein-binding interfaces by using 38 optimally chosen properties. The optimal combination of features was selected from a set of 324 novel structural neighborhood properties by a two-step feature selection method consisting of a random forest algorithm and a sequential backward elimination method. We evaluated the performance of PredHS using a benchmark of 265 alanine-mutated interface residues (Dataset I) and a trimmed subset (Dataset II) with 10-fold cross-validation. Compared with the state-of-the art approaches, PredHS achieves a significant improvement on the prediction quality, which stems from the new structural neighborhood properties, the novel way of feature generation, as well as the selection power of the proposed two-step method. We further validated the capability of our method by an independent test and obtained promising results.