Predicting the potency of hERG K⁺ channel inhibition by combining 3D-QSAR pharmacophore and 2D-QSAR models.

Blockade of the hERG K(+) channel has been identified as the most important mechanism of QT interval prolongation and thus inducing cardiac risk. In this work, an ensemble of 3D-QSAR pharmacophore models was constructed to provide insight into the determinants of the interactions between the hERG K(+) channel and channel inhibitors. To predict hERG inhibitory activities, the predicted values from the ensemble of models were averaged, and the results thus obtained showed that the predictive ability of the combined 3D-QSAR pharmacophore model was greater that those of the individual models. Also, using the same training and test sets, a 2D-QSAR model based on a heuristic machine-learning method was developed in order to analyze the physicochemical characters of hERG inhibitors. The models indicated that the inhibitors have certain key inhibitory features in common, including hydrophobicity, aromaticity, and flexibility. A final model was developed by combining the combined 3D-QSAR pharmacophore with the 2D-QSAR model, and this final model outperformed any other individual model, showing the highest predictive ability and the lowest deviation. This model can not only predict hERG inhibitory potency accurately, thus allowing fast cardiac safety evaluation, but it provides an effective tool for avoiding hERG inhibitory liability and thus enhanced cardiac risk in the design and optimization of new chemical entities.