Predicting DNA-Binding Sites by Exploring the Distribution of Atom Groups around the Surface

摘要

DNA-binding proteins perform various functions in the cells. Determining the structures of protein-DNA complexes using experimental methods are hindered by many obstacles. Thus, computational methods for predicting DNA-binding sites on protein structures are needed to elucidate the mechanism of protein-DNA interactions. In this study, we divided atoms of amino acid residues into 14 groups and used a vector consisting of the distribution of these atom groups to describe the characteristics of protein surface around an amino acid. We then trained a Random Forest method to predict DNA-binding sites on protein surface. The predictions were then refined using a post-processing procedure based on the clustering of DNA-binding residues on the surface. The method achieved an accuracy of 80.8% when evaluated using 10-fold cross-validation. The results show that the distribution of different types of atoms around the surface provides sufficient structural information for predicting DNA-binding sites on protein structures.