A computationally based identification algorithm for estrogen receptor ligands: Part 2. Evaluation of a hER alpha binding affinity model

The objective of this study was to evaluate the capability of an expert sys tem described in the previous paper (S. Bradbury et al., Toxicol. Sci. 58, 253-269) to identify the potential for chemicals to act as ligands of mamma lian estrogen receptors (ERs). The basis of the expert system was a structu re activity relationship (SAR) model, based on relative binding affinity (R BA) values for steroidal and nonsteroidal chemicals derived from human ER a lpha (hER alpha) competitive binding assays. The expert system enables cate gorization of chemicals into (RBA ranges of < 0.1, 0.1 to 1, 1 to 10, 10 to 100, and >150% relative to 17 beta -estradiol. In the current analysis, th e algorithm was evaluated with respect to predicting RBAs of chemicals assa yed with ERs from MCF7 cells, and mouse and rat uterine preparations. The b est correspondence between predicted and observed RBA ranges was obtained w ith MCF7 cells. The agreement between predictions from the expert system an d data from binding assays with mouse and rat ER(s) were less reliable, esp ecially for chemicals with RBAs less than 10%. Prediction errors often were false positives, i.e., predictions of greater than observed RBA values. Wh ile discrepancies were likely due, in part, to species-specific variations in ER structure and ligand binding affinity, a systematic bias in structura l characteristics of chemicals in the hER alpha training set, compared to t he rodent evaluation data sets, also contributed to prediction errors. Fals e-positive predictions were typically associated with ligands that had shie lded electronegative sites. Ligands with these structural characteristics w ere not well represented in the training set used to derive the expert syst em. Inclusion of a shielding criterion into the original expert system sign ificantly increased the accuracy of RBA predictions. With this additional s tructural requirement, 38 of 46 compounds with measured RBA values greater than 10% in hER alpha, MCF7, and rodent uterine preparations were correctly categorized. Of the remaining 129 compounds in the combined data sets, RBA values for 65 compounds were correctly predicted, with 47 of the incorrect predictions being false positives. Based upon this exploratory analysis, t he modeling approach, combined with a high-quality training set of RBA valu es derived from a diverse set of chemical structures, could provide a credi ble tool for prioritizing chemicals with moderate to high ER binding affini ty for subsequent in vitro or in vivo assessments.