In Silico Prediction of Caco-2 Cell Permeability by a Classification QSAR Approach.

In the present study, 21 validated QSAR models that discriminate compounds with high Caco-2 permeability (Papp ≥8×10(-6)  cm/s) from those with moderate-poor permeability (Papp <8×10(-6)  cm/s) were developed on a novel large dataset of 674 compounds. 20 DRAGON descriptor families were used. The global accuracies of obtained models were ranking between 78-82 %. A general model combining all types of molecular descriptors was developed and it classified correctly 81.56 % and 83.94 % for training and test sets, respectively. An external set of 10 compounds was predicted and 80 % was correctly assessed by in vitro Caco-2 assays. The potential use of the final model was evaluated by a virtual screening of a human intestinal absorption database of 269 compounds. The model predicted 121 compounds with high Caco-2 permeability and the 90 % of them had high values of human intestinal absorption (HIA≥80). This study provides the most comprehensive database of Caco-2 permeability and evidenced the utility of the combined methodology (in silico+in vitro) in the prediction of Caco-2 permeability. It suggests that the present methodology can be used in the design of large libraries of compounds with appropriate values of permeability and to perform virtual screening in the early stages of drug development.