You are here:
PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: MODEL I
Citation:
Serafimova, R., O. G. Mekenyan, AND R. Serafimova. PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: MODEL I. Presented at 10th International Workshop on Quantitative Structure-Activity Relationships (QSARs) in Environmental Sciences, Ottawa, Ontario, Canada, May 25-29, 2002.
Description:
A Common Reactivity Pattern (COREPA) model, based on consideration of multiple energetically reasonable conformations of flexible chemicals was developed using a training set of 232 rat estrogen receptor (rER) relative binding affinity (RBA) measurements. The training set included 100 active and 132 inactive ER ligands using 0.001% RBA, relative to estradiol as the cutoff for activity. The model developed using this training set successfully predicted 98% of training set chemicals to be active. A rule file was developed and coded into a decision tree format used to predict the ER binding affinity of 6757 high production volume chemicals. The model generated RBA predictions for 3769 chemicals, from the initial 6757 dataset for which structural definition was available, after the elimination of chemicals of MW < 81and those with no cycle fragments. Only about 10% of the thousands of chemicals wre predicted to have estrogenic activity. Of these - 300 chemicals approximately 8% were predicted to bind with affinities of 10% to > 100% relative to 17B-estradiol. Half of the active chemicals are predicted to bind with from 0.1% to > 100% RBA, with the rest having predicted RBA of 0.001% to 0.1%. An approach for using additional measured RBA data to assess the accuracy of model predictions is presented in the context of active vs. inactive chemicals, as well as comparisons of predicted and measured data within six order of magnitude activity ranges. This abstract does not necessarily reflect USEPA policy.