You are here:
A COMPUTATIONALLY-BASED IDENTIFICATION ALGORITHM FOR POTENTIAL ESTROGEN RECEPTOR LIGANDS, PART II. AN EVALUATION OF A HUMAN RECEPTOR-BASED MODEL
Citation:
Mekenyan, O. G., V. Kamenska, P. K. Schmieder, G T. Ankley, AND S P. Bradbury. A COMPUTATIONALLY-BASED IDENTIFICATION ALGORITHM FOR POTENTIAL ESTROGEN RECEPTOR LIGANDS, PART II. AN EVALUATION OF A HUMAN RECEPTOR-BASED MODEL. TOXICOLOGICAL SCIENCES 58:270-281, (2000).
Description:
The objective of this study was to evaluate the capability of an expert system described in the previous paper (Bradbury et al., 2000; Toxicol. Sci.) to identify the potential for chemicals to act as ligands of mammalian estrogen receptors (ERs). The basis of that algorithm was a structure-activity relationship (SAR) model derived from the interaction of steroidal and non-steroidal chemicals with the human ER-a in competitive binding assays. The expert system enables categorization of chemicals into relative binding affinity (RBA) ranges of <0.1, 0.1 to 1.1 to 10, 10 to 100, and > 150%, relative to B-estradiol. In the current analysis, we assessed the algorithm with respect to predicting RBAs of chemicals in assays with ERs from MCF7 cells, and mouse and rat uterine preparations. The best correspondence between output of the predictive system and empirical data was for the MCF7 cells. The agreement between predictions from the expert system and data from binding assays with mouse and rat ER9s) was not as robust, especially for RBAs less than 10%. Part of this likely was due to species-specific variations in ER structure and ligand binding affinity; however, at least some component of the variability in predicted versus measured RBA ranges appeared to be due to a systematic bias in structural characteristics of chemicals in the human ERa versus rodent data sets. Specifically, observed "misidentifications" often were false positives, i.e., prediction of greater affinity for the ER than was actually measured. This was a common occurrence for ligands with shielded electronegative sites, which were not well represented in the training set used to derive the initial SAR model. Inclusion of a criterion based on this structural characteristic into the original expert system significantly increased accuracy of RBA predictions. For compounds with measured RBA values greater than 10% in hERa, MCF7, and mouse and rat uterine preparations, 38 of 46 compounds were correctly - - -