A New Neural Network for B-Turn Prediction: the Effect of Site-Specific Amino Acid Preference

The prediction of β-turn, despite the observation that one out of four residues in protein belongs to this structure element, has attracted considerably less attention comparing to secondary structure predictions. Neural network machine learning is a popular approach to address such a problem of structural bioinformatics. In this paper, we describe a new neural network model for β-turn prediction that accounts for site-specific amino acid preference, a property ignored in previous training models. We showed that the statistics of amino acid preference at specific sites within and around a β-turn is rather significant, and incorporation of this property helps improve the network performance. Furthermore, by contrasting with a previous model, we revealed a deficiency of not incorporating this site-specific property in previous models.