One of the major reasons for late-stage failure of drug candidates is due to problems uncovered in pharmacokinetics during clinical trials. There is now a general consensus for earlier consideration of these effects in the drug discovery process. Computer-aided design technology provides us with tools to develop predictive models for such pharmacokinetic properties. Among these tools, we focus on pharmacophore modeling techniques in this article. Pharmacophore models that are reported for various cytochrome P450 (CYP) enzymes are reviewed for the isoenzymes CYP1A2, 2B6, 2C9, 2C19, 2D6, 2E1, and 3A4. In addition pharmacophore models for related metabolic processes through CYP19 (aromatase), CYP51 (14α-lanosterol demethylase), PXR (pregnane X-receptor), and finally for human intrinsic clearance are also reviewed. The models reported by various scientists are schematically represented in the figures in order to visually demonstrate their similarities and differences. The models developed by different researchers or sometimes even by the same research group for different sets of ligands, provide a clear picture of the challenges in coming up with a single model with good predictive values. One of the main reasons for this challenge is related to relatively large size of the active sites and flexibility of the CYP isoenzymes, which results in multiple binding sites. We propose development of multiple- diverse pharmacophore models for each binding mode (as opposed to a single predictive model for each CYP isoenzyme). After scoring and prioritization of the models, we propose the use of a battery of pharmacophore models for each CYP isoenzyme binding mode to computationally obtain a P450 interaction profile for drug candidates early in the drug development cycle, when decisions on their fate can be made before incurring the costs of synthesis and testing.
Keywords: 3D-QSAR, ADME profiling, drug-drug interactions, cytochrome P450, CYP isoenzymes, metabolic pathways, pharmacokinetic predictions, pharmacophore modeling, predictive ADME/Tox, quantitative pharmacophore models, QSAR.