Application of Neural Network for Locating the Binding Sites in HCV Protein Complex

Hepatitis C virus infection remains a public health problem within an international scope, and much efforts has been devoted to understanding the interaction within HCV and human protein complex. Among several attempts, identification of binding site and prediction of critical residue contribute significantly to the function of a protein, and can narrow the search space required by docking algorithms to predict the structures of complexes. In the paper, we combined a neural network model with surface patch analysis, which includes residue binding propensity, sequence entropy, hydrophobicity and conservation, to predict protein-protein binding sites within a comprehensive set of 18 HCV protein complexes retrieved from protein data bank. We correctly predicted the binding sites with an average performance of 91.39%. Extended to a set of 108 virus and human protein complexes, our method still achieved a preferable result with an average performance of 91.61%. Future research on interaction within HCV proteins or with other proteins may benefit from these results.