Computational classification models for predicting the interaction of compounds with hepatic organic ion importers

Hepatic transporters, a major determinant of pharmacokinetics, have been used to profile drug properties like efficacy. Among hepatic transporters, importers alter the concentration of the drug by facilitating the transport of a drug into a cell. Despite vast pharmacokinetic studies, the interacting mechanisms of the importers with its substrates or inhibitors are not well understood. Hence, we developed compound binary classification models of whether a compound is binder or nonbinder to a hepatic transporter with experimental data of 284 compounds for four representative hepatic importers, OATP1B1, OATP1B3, OAT2, and OCT1. Support Vector Machine (SVM) along with Genetic Algorithm (GA) was used to construct the classification models of binder versus nonbinder for each target importer. To construct the models, we prepared two data sets, a training data set from Fujitsu database (284 compounds) and an external validation data set from ChEMBL database (1738 compounds). Since an experimental classification criterion between binder and nonbinder has some ambiguity, there is an intrinsic limitation to expect high predictability of the binary classification models developed with the experimental data. The predictability of the classification models calculated with external validation sets were obtained as 77.72%, 84.31%, 84.21%, and 76.38 for OATP1B1, OATP1B3, OAT2, and OCT1, respectively.