VHSE descriptors were employed by using GA-PLS to establish QSAR models of 102 HLA-A*0201 restrictive CTL epitopes for major histocompatibility complex or human leukocytoantigen. After removing 3 outliers, total 99 samples were finally used to develop QSAR models. According to the results of cross-validation and those of external validation on 50 test samples, an optimal PLS model (A＝2) was derived whose R², Q² (leave one out) and were 0.755, 0.621 and 0.680 respectively. From the results of QSAR studies, it can be seen that hydrophobic properties at positions 1, 2, 7, 8 and 9, steric properties at positions 1 and 2 together with electronic properties at positions 4 and 6, were close related to the binding affinities of CTL epitopes.

Key words: cytotoxic T lymphocyte, epitope, quantitative structure-activity relationship, genetic algorithm, partial least squares