Comparing with experimental HTS, virtual screening (VS) is more efficient and economical in identifying biologically active molecules. Here we present a genetic algorithm (GA) guided pharmacophore model optimization program, where a pharmacophore model is represented by a chromosome that encodes the coordinates and tolerance radii of pharmacopheric features. The training data consists of 45 estrogen receptor alpha (ERa) agonists identified from a cell-based qHTS assay and 79 inactives that have the highest structural resemblance to the agonists. The original model, containing up to 5 chemical features and 354 exclusive volumes to define the shape of the ERa ligand binding site, recognized 28 true positives and 40 false positives. The optimized model was able to retrieve 24 true positives that were the most potent in the training set without any false positives. Applying the optimized model to a 10K compound collection, 47 out of the 292 actives from the screen were predicted by the model as active, with 49 false positives. A large fraction of the false negatives are esters of known ERa agonists, which are readily hydrolyzed in cell to release the active form. The impact of fitness functions and the definition of chemical features on the performance of the optimized models will be elaborated. Future direction of this project will be discussed.

Genetic algorithm guided optimization of pharmacophore models for estrogen receptor agonists

244th National Meeting (2012)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.