In silico predictions of ADME-Tox properties: drug absorption.

The accurate prediction of the in-vivo pharmacokinetics of a new potential drug compound based on only its virtual structure is the ultimate goal of in-silico ADME-Tox property modeling. A comprehensive review is made on recent studies concerning the A (absorption) in ADME-Tox, i.e. the in-silico modeling of both Caco-2 permeability and human intestinal absorption. The data sets used, the descriptors selected to build the models, the variable selection methods, the modeling techniques and the performed model validation are critically discussed. It was concluded that reliable models which improve the success rate of compound selection and drug development are still lacking. Limiting the quality of the models are, for instance, inappropriate compilation of data sets, lack of an appropriate outlier detection and unrepresentativeness of training and test sets for the data population. The definition of some best practices or guidelines for the different steps of the modeling procedure might improve the predictions and make the procedure uniform, i.e. "standard tools" in drug development would become available.