Contrast Pattern Mining with Gap Constraints for Peptide Folding Prediction

1 In this paper, we propose a peptide folding prediction method which discovers contrast patterns to differentiate and predict peptide folding classes. A contrast pattern is defined as a set of sequentially associated amino acids which frequently appear in one type of folding but significantly infrequent in other folding classes. Our hypothesis is that each type of peptide folding has its unique interaction patterns among peptide residues (amino acids). The role of contrast patterns is to act as signatures or features for prediction of a peptide’s folding type. For this purpose, we propose a two phase peptide folding prediction framework, where the first stage is to discover contrast patterns from different types of contrast datasets, followed by a learning process which uses all discovered patterns as features to build a supervised classifier for folding prediction. Experimental results on two benchmark protein datasets will indicate that the proposed framework can outperform simple secondary structure prediction based approaches for peptide folding prediction.