An accurate prediction of the pharmacokinetic properties of orally administered drugs is of paramount importance
in pharmaceutical industry. Caco-2 cell permeability is a well established parameter for assessing the drug absorption
profiles of lead molecules. Due to the restrictions on animal testing, prohibitive in situ models and ethical issues, the
development of predictive models is essential. Genetic programming (GP) is an artificial intelligence (AI)-based exclusively
data driven modeling paradigm. Given an example input-output data, it searches and optimizes, both the structure
and parameters of a well fitting linear/non-linear input-output model. Despite this novelty, GP has not been widely exploited
in drug design. Accordingly, in this study we propose a GP based approach for the in silico prediction of Caco-2
cell permeability using a diverse set of molecules. The predictions yielded a high magnitude for the training and test set
correlation coefficient with low RMSE, indicating accurate Caco-2 permeability prediction and generalization performance
by the GP model. The predictions were better or comparable to artificial neural networks (ANN) and support vector
regression (SVR) methods. The GP based modeling approach illustrated will find diverse applications in (QSAR, QSPR
and QSTR) modeling for the virtual screening of large libraries.
ADME modeling, Caco-2 cell permeability, genetic programming, MLP, SVR.