The transporter associated with antigen processing (TAP) plays a crucial role in the antigen-processing pathway mediated by human leukocyte antigen(HLA) class Ⅰ molecules. TAP binding affinity of peptide can affect T-cell epitope selection significantly, so prediction of TAP binding affinity as well as selection specificity research has become a hot spot in computational immunology recently. Compared to the existing methods with little physicochemical meanings, in this paper, VHSE(principal component score vector of hydrophobic, steric, and electronic properties), a novel set of descriptors based on the physical and chemical characteristics of 20 nature amino acids, was used to characterize 613 nonamer peptides of known affinity to TAP. Then, support vector machine(SVM) combined with multiple stepwise regression(MSR) was adopted to establish prediction models of TAP binding affinity of peptide. An optimal SVM model with linear kernel was obtained, of which R2, Q2 and Rext2 were 0.7386, 0.7270 and 0.6057, respectively. The results show that electronic, steric, and hydrophobic properties of the amino acid sites are closely related to TAP binding affinity and substrate specificity, especially for electrical property. The amino acids at P1, P2, P7 and P9 sites of antigenic peptides have the most impact on TAP binding affinity, while those at P3, P4, P5 and P6 sites have less impact and amino acid at P8 site has no impact. According to the prediction results of single point mutated antigenic peptides by the optimal linear SVM model together with the contribution weights of selected variables, substrate specificity of TAP was summarized.
