Support vector machine and pharmacophore-based prediction models of multidrug-resistance protein 2 (MRP2) inhibitors

Overexpression of multidrug-resistance protein 2 (MRP2) is one of the main causes that lead the curative effect reduction of many drugs, particularly anticancer drugs. Development of MRP2 inhibitors is the aim to overcome multidrug resistance due to MRP2. In this study, computational prediction models of MRP2 inhibitors have been developed by using support vector machine (SVM) and pharmacophore modeling method. For the SVM model, the overall prediction accuracy is 82.9% for the training set (257 compounds) and 77.1% for the independent test set (61 compounds). And 16 descriptors have been used in the SVM modeling; but from which it is difficult to get understanding about the action mechanism. The established pharmacophore model Hypo1 consists of two hydrogen bond acceptors and one hydrophobic feature. With the use of Hypo1, 78.1% of MRP2 inhibitors and 69.6% non-inhibitors can be predicted correctly. The overall prediction accuracy is 73.9%. Although the prediction accuracy of the pharmacophore model is lower than that of SVM model, it gives a clear picture of chemical features necessary for the MRP2 inhibitors. Taken together, the SVM model is capable of predicting MRP2 inhibitors with considerable good accuracy. But the gain of action mechanism related information needs the help of pharmacophore model.