You are here:
PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: MODEL II
Citation:
Serafimova, R., Z. Nedjalkova, O. G. Mekenyan, P. K. Schmieder, AND Z. Nedyalkova. PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: MODEL II. Presented at 10th International Workshop on Quantitative Structure-Activity Relationships (QSARs) in Environmental Sciences, Ottawa, Ontario, Canada, May 25-29, 2002.
Description:
The training set used to derive a common reactivity pattern (COREPA) model for estrogen receptor (ER) binding affinity in Model I (see Abstract I in this series) was extended to include 47 rat estrogen receptor (rER) relative binding affinity (RBA) measurements in addition to the 232 in the initial training set. This resulted in a new training set of 144 active and 135 inactive ER ligands (>0.001% RBA defining activity) which included additional substituted steroids and hydroxylated PCBs, rat estrogen receptor (rER) relative binding affinity (RBA) measurements. The new training set was used to develop Model II, which predicted the training set accurately with 1.4% false negatives and 6% false positives. The new model was used to predict ER binding affinity of 6757 chemicals (Universe I). The expert system (rule file and decision tree) developed from Model II identified more active ER ligands (> 350) than did the initial model (-300) which used a smaller, less diverse training set, with fewer active ER ligands. As with the initial model, when chemicals were placed in order of magnitude RBA ranges, almost half of the chemicals predicted to be active were found in the two lowest RBA ranges, having binding affinities 0.001% that of estradiol, similar to Model I. The data shown compares and contrasts the two models developed as well as the results of application of each model to the screening of large chemical datasets. Additionally the concept of conformer multiplication as applied to model development and performance will be discussed. This abstract does not necessarily reflect USEPA policy.