A Logistic Regression Approach for Identifying Hot Spots in Protein Interfaces

Abstract

Protein–protein interactions occur when two or more proteins bind together, often to carry out their biological function. A small fraction of interfaces on protein surface found providing major contributions to the binding free energy are referred as hot spots. Identifying hot spots is important for examining the actions and properties occurring around the binding sites. However experimental studies require significant effort; and computational methods still have limitations in prediction performance and feature interpretation.

In this paper we describe a hot spots residues prediction measure which provides a significant improvement over other existing methods. Combining 8 features derived from accessibility, sequence conservation, inter-residue potentials, computational alanine scanning, small-world structure characteristics, phi-psi interaction, and contact number, logistic regression is used to derive a prediction model. To demonstrate its effectiveness, the proposed method is applied to ASEdb. Our prediction model achieves an accuracy of 0.819, F1 score of 0.743. Experimental results show that the additional features can improve the prediction performance. Especially phi-psi has been found to give important effort. We then perform an exhaustive comparison of our method with various machine learning based methods and those previously published prediction models in the literature. Empirical studies show that our method can yield significantly better prediction performance.