Comparison of two quantitative matrices for in silico prediction of major histocompatibility complex class I peptide loading

Abstract

Epitope prediction using in silico methods serves as an efficient and cost-effective method for vaccine design. A mechanism by which vaccines elicit immune responses involves the activation of cytotoxic T lymphocytes (CTL), which requires that peptide epitopes are sufficiently loaded onto major histocompatibility complex (MHC) Class I molecules. A major consideration included physical interactions between the amino acids comprising the peptide epitope and the specific MHC Class I protein. In this paper, the predictive performances of two published quantitative matrices, namely, the re-scaled Miyazawa-Jernigan (MJ) or Betancourt-Thirumalai (BT), and quasi-chemical pair correlation (QC), were evaluated, showing differential allele-specific predictive performance for loading. In addition, since QC matrix has a higher average AUC than BT matrix, we recommend the use of QC matrix as a scoring function for peptide loading classification and regression algorithms.