In silico prediction of drug properties

Drug design has become inconceivable without the assistance of computer-aided methods. In this context in silico was chosen as designation to emphasize the relationship to in vitro and in vivo testing. Nowadays, virtual screening covers much more than estimation of solubility and oral bioavailability of compounds. Along with the challenge of parsing virtual compound libraries, the necessity to model more specific metabolic and toxicological aspects has emerged. Here, recent developments in prediction models are summarized, covering optimization problems in the fields of cytochrome P450 metabolism, blood-brain-barrier permeability, central nervous system activity, and blockade of the hERG-potassium channel. Aspects arising from the use of homology models and quantum chemical calculations are considered with respect to the biological functions. Furthermore, approaches to distinguish drug-like substances from nondrugs by the means of machine learning algorithms are compared in order to derive guidelines for the design of new agents with appropriate properties.