Predicting peptide bond conformation using feature selection and the Naive Bayes approach

Distinguishing cis peptide bonds from trans isomers in protein sequences facilitates the exploration of protein structures and functions. In this study, we evaluated the effect of a large and informative feature vector, towards the reliable prediction of peptide bond conformation between any two amino acids. We used multiple sequence alignment, secondary structure information, real valued solvent accessibility predictions for each amino acid and physicochemical properties of the surrounding residues. A three stage schema was developed, comprising of feature extraction, feature selection and peptide bond classification between any two amino acids. We also explored the performance achieved when using the full feature vector without performing feature selection. The best discriminating ability was achieved using a Naive Bayes classifier, combined with wrapper feature selection. The proposed approach yielded prediction accuracy 86%, sensitivity 82% and specificity 90% in discriminating cis and trans peptide bond conformations.