|
Health
and Human Services | Food and Drug Administration
| NCTR | | EDKB Home | Database | Endocrine Disruptors | Experimental Program | Models | Publications |
 |
|
|
Health
and Human Services | Food and Drug Administration
| NCTR | | |
|
We have developed a broad range of models that include rule-based, pharmacophore, chemometric and QSAR techniques, and integrated the suite of the models together in a complementary manner to provide a system for screening and prioritizing for testing. Critical to the robust validity of these models is an ER- alpha competitive binding dataset from our own assays [Blair, 2000 #216] used for model calibration and testing. The chemicals in the dataset were selected for their chemical structure diversity and range of activity, criteria essential to develop robust and valid models, but to our knowledge lacking for calibration of previously developed models. Recently available crystallographic data [Shiau, 1998 #73; Tanenbaum, 1998 #247; Brzozowski, 1997 #11] for ER agonists and antagonists figured prominently in guiding model development.
The circular process figure below depicts our model development process that has thoroughly integrated a supporting experimental program, and incorporates data validation and chemical structure diversity analysis. The upper three boxes in the figure show the steps that have been used to develop initial models. In this case, available data from the literature were used to develop models. Cross validation and testing on a test set provide validation. Unfortunately, available data were insufficient to account fully for structural diversity and to provide a diverse test set capable of truly challenging and validating the model, or elucidating the model's weakness and potential hypotheses for improvement.
The process depicted in the figure is circular and incorporates aspects to provide a training set designed to produce a robust model. Experimental validation techniques [Blair, 2000 #216] were used to evaluate each data point for potential experimental errors, potential for multiple or confounding mechanisms of action, and potential for contamination by impurities (particularly active impurities). Models for prediction and for representation of structural diversity were developed and used in concert with the conduct of the experimental assays providing a measure of assurance that chemicals span the required range of chemical structure descriptor space. Benefits from the integration of the experimental and modeling efforts include:
