Drugs that inhibit the human ether-a-go-go related gene are at risk to lead to a prolongation of the QT interval or torsade de pointes in the worst case.  Therefore, it is important to devise a model to predict the hERG liability at the early stage of drug discovery.  A pharmacophore ensemble is constructed from a number of pharmacophore hypotheses to address the plasticity of hERG protein while interacting with various compounds; and is subject to regression by support vector machine to generate the final model.  The SVM-based model performed better than any of pharmacophore candidates in the ensemble and yielded the correlation coefficients of 0.98 and 0.97 for the training set and test set, respectively, suggesting that this is a plausible in silico model to predict the hERG liability of novel compounds.