SVM-Based Approach for Predicting DNA-Binding Residues in Proteins from Amino Acid Sequences

Protein-DNA interactions are vitally important in a wide range of biological processes such as gene regulation and DNA replication and repair. We predict DNA-binding residues in proteins from amino acid sequences by support vector machine (SVM) with a novel hybrid feature which incorporates evolutionary information of amino acid sequences and four physical–chemical properties, including the side chain pKa value, hydrophobicity index, molecular mass and lone electron pairs of amino acids. The classifier achieves 79.12% total accuracy with 74.19% sensitivity and 79.20% specificity, respectively. Moreover, an alternative classifier using random forest (RF) is also constructed. Further analysis proves that the hybrid feature shows obvious contribution to our excellent prediction performance, and the evolutionary information contributes most to the prediction improvement.