Predicting the toxic potential of drugs and chemicals in silico: A model for the peroxisome proliferator-activated receptor gammaAuthors: Angelo Vedani, Anne-Vérène Descloux, Morena Spreafico and Beat Ernst
Journal: Toxocology Letters
Biographics Laboratory 3R, Friedensgasse 35, 4056 Basel, Switzerland and
Department of Pharmaceutical Sciences, University of Basel, Klingelbergstrasse 50, 4056 Basel, Switzerland.
Poor pharmacokinetics, side effects and compound toxicity are frequent causes of latestage failures in drug development. A safe in silico identification of adverse effects triggered by drugs and chemicals would therefore be highly desirable as it not only bears economical potential but also spawns a variety of ecological benefits: sustainable resource management, reduction of animal models and possibly less risky clinical trials as in silico studies are typically based on human proteins. In the recent past, our laboratory has developed a 6D-QSAR concept and validated a series of "virtual test kits" based on the aryl hydrocarbon, estrogen, androgen, thyroid, and glucocorticoid receptor as well as on the enzyme cytochrome P450 3A4. The test kits were trained using a representative selection of 610 substances and validated with 188 compounds different therefrom. These models were subsequently compiled into a database for the virtual screening of drugs and environmental chemicals. In this account, we report the validation of a model for the peroxisome proliferator-activated receptor γ (PPAR γ). Its receptor surrogate is based on the experimental structure of the protein and 95 tyrosine-based compounds. The simulation reached a cross-validated r2 = 0.832 (75 training ligands) and yielded a predictive r2 = 0.723 (20 test compounds). The model was challenged by a series of scramble tests as well as with the prediction of a few structurally different compounds.
Quasar model of the peroxisome proliferator-activated receptor γ with the most potent ligand bound.
Comparison of experimental and predicted Ki values. Green dots mark the ligand molecules of the training set, red dots identify those belonging to the test set; dashed lines mark a factor of five and ten from the experimental value, respectively.