Improving the prediction of HLA class I-binding peptides using a supertype-based method.

Abstract: The computational prediction of peptides that bind to major histocompatibility complex (MHC) molecules has practical importance for the development of epitope-based vaccines. The performance of the prediction methods depends on the verified peptides. However, the available peptide datasets of most alleles contain significant biases. An investigation to the effect of the peptides in the training dataset on the performance of the generated model indicated that there was a discrepancy between the classification of binders from biological data and classification of binders from super-motif-sharing peptides, which was induced by the non-motif-containing peptides. Most human MHC (called HLA) class I molecules could be assigned to supertypes based on their overlapping peptide-binding specificities, therefore, we proposed a supertype-based method for the modeling of the HLA class I-peptide binding: candidates of peptides binding to alleles in a given supertypes were screened using the super-motifs, and then the peptides binding to specific allele in the supertype were predicted by the model trained on the super-motif-sharing peptides. The efficacy of this supertype-based method was examined in two matrix-based methods and one machine learning method for 20 alleles in HLA supertypes A1, A2, A3, A24, B44 and B7. Evaluations on several benchmark datasets indicated that the supertype-based method achieved remarkable success in improving the prediction of HLA-binding peptides. [Copyright &y& Elsevier]