You are here:
PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: A COMPARISON OF THREE MODELS
Citation:
Serafimova, R., T. Pavlov, O. G. Mekenyan, AND P. K. Schmieder. PREDICTING ER BINDING AFFINITY FOR EDC RANKING AND PRIORITIZATION: A COMPARISON OF THREE MODELS. Presented at 10th International Workshop on Quantitative Structure-Activity Relationships (QSARs) Environmental Sciences, Ottawa, Ontario, Canada, May 25-29, 2002.
Description:
A comparative analysis of how three COREPA models for ER binding affinity performed when used to predict potential estrogen receptor (ER) ligands is presented. Models I and II were developed based on training sets of 232 and 279 rat ER binding affinity measurements, respectively. A third model was developed using data in the initial models plus an additional 35 chemicals whose ER binding affinities were measured additional mammalian species (human or mouse). Where the same chemical was measured in multiple species the highest RBA value recorded was used in an attempt to minimize false negative predictions at the possible expense of false positives. The additional data was added to increase training set structural diversity. Details of the mammalian model and it's ability to predict the training set is presented here and compared with similar analyses presented in posters 1 and 2 of this series. The performance of this final model for prediction of ER affinity of the 6757 chemicals is similar to that of earlier models in total numbers of chemicals predicted active, however almost twice as many chemicals are predicted to have binding affinities from 10% to > 100% that of estradiol. This could be explained by the different structural diversity introduced by the additional chemicals in mammalian training set as well as by the rearrangement of chemicals in different RBA ranges of mammalian training set, due to the algorithm used for determining RBAs over different species. An analysis of how many and what types of chemicals are predicted to have binding affinities in each of the six order of magnitude binding affinity ranges is presented for each model and analyzed in the context of the chemical composition and potencies in each of the training sets. This abstract doe not necessarily reflect USEPA policy.