Two-step models to predict binding affinity of chemicals to the human estrogen receptor α by three-dimensional quantitative structure-activity relationships (3D-QSARs) using receptor-ligand docking simulation

Binding of chemicals to the estrogen receptor (ER) is known to be a key mode of action of endocrine disruption effects. In this study, combined quantitative structure-activity relationship (QSAR) models from discriminant and multilinear regression (MLR) analyses, termed a two-step model, were developed. These were used to predict the binding potency to human ER alpha of four chemical groups, namely alkylphenols, phthalates, diphenylethanes and benzophenones. These groups are considered to be important chemical classes of ER-binders. The descriptors investigated were calculated following the simulation of docking between the receptor and ligand. Discriminant analysis in the first step of a two-step model was applied to distinguish binders from non-binders. It had a concordance, following leave-one-out (LOO), of greater than 87% for all chemical classes. Binders were defined as chemicals whose IC50 was reliably measured in a competitive binding assay. The MLR analysis in the second step was performed for the quantitative prediction of the binding affinity of chemicals that were previously discriminated as binders. The q(2) values for alkylphenols and diphenylethanes were 0.75 and 0.74, respectively. However good MLR relationships were not obtained for phthalates and benzophenones as the observed binding affinities of chemicals in these categories were weak and in a too narrow range.