Abstract: The binding affinity model of a peptide-major histocompatibility complex (MHC) was studied using the methods of partial least squares (PLS) and amino acid structure descriptors. Using these models, the predictive sets of 805 peptides are predicted for the binding affinity of the HLA-A*0201 complex, and the predictive accuracies of the two models are 66.8% and 65.5%, respectively. To test the robustness of the models, the outliers are eliminated, and it is proved that the robustness of the models is good. This shows that the models using the PLS and amino acid structure descriptors are feasible on predicting of CTL (cytotoxicy T lymphocyte) epitopes.

Key words: CTL epitopes, Amino acid structure descriptors, QSAR, Partial least squares method