The overexpression of ABCC2/MRP2, an ATP-binding cassette transporter, contributes to multidrug resistance in cancer cells. In this study, a quantitative structure-activity relationship (QSAR) analysis on ABCC2 inhibitors has been carried out, aiming to establish a computational prediction model for ABCC2 modulators. Seven classification models and two regression models were built by SONNIA 4.2, and two other regression models were built by MOE 2008.10 based on a data set comprising 372 compounds collected from 16 relevant publications. The CPG-C iABCC2 model for classifying ABCC2 inhibitors has total accuracy of 0.88 and Matthews correlation coefficient MCC = 0.75. The CPG-C iEG model for classifying ABCC2 inhibitors (substrate EG: β-estradiol 17-β-D-glucuronide) has total accuracy of 0.91 and MCC = 0.82. The regression model PLS EG-IC50 for predicting ABCC2 inhibitors (substrate EG) gave root-mean-square error RMSE = 0.26, Q2 = 0.73 and [Formula: see text]. The regression model PLS CDCF-IC50 for predicting ABCC2 inhibitors [substrate CDCF: 5(6)-carboxy-2',7'-dichlorofluorescein] gave RMSE = 0.31, Q2 = 0.74 and [Formula: see text]. Four 2D-QSAR models were applied to 1661 compounds, with results indicating 369 compounds having the ability to reverse the efflux of both EG and CDCF by ABCC2, 152 among them having IC50 < 100 µM.
Keywords: ABCC2; Classification; Inhibitor; QSAR; Regression; Virtual screening.