Indexing molecules for their hERG liability

•Early prediction of drug liability to hERG K+ channel is highly important.•ISE algorithm was applied for indexing molecules for their hERG liability.•A highly discriminative and robust model is proposed.•The proposed model could be implemented in screening and for hit/lead optimization.

The human Ether-a-go-go-Related-Gene (hERG) potassium (K+) channel is liable to drug-inducing blockage that prolongs the QT interval of the cardiac action potential, triggers arrhythmia and possibly causes sudden cardiac death. Early prediction of drug liability to hERG K+ channel is therefore highly important and preferably obligatory at earlier stages of any drug discovery process. In vitro assessment of drug binding affinity to hERG K+ channel involves substantial expenses, time, and labor; and therefore computational models for predicting liabilities of drug candidates for hERG toxicity is of much importance. In the present study, we apply the Iterative Stochastic Elimination (ISE) algorithm to construct a large number of rule-based models (filters) and exploit their combination for developing the concept of hERG Toxicity Index (ETI). ETI estimates the molecular risk to be a blocker of hERG potassium channel. The area under the curve (AUC) of the attained model is 0.94. The averaged ETI of hERG binders, drugs from CMC, clinical-MDDR, endogenous molecules, ACD and ZINC, were found to be 9.17, 2.53, 3.3, −1.98, −2.49 and −3.86 respectively. Applying the proposed hERG Toxicity Index Model on external test set composed of more than 1300 hERG blockers picked from chEMBL shows excellent performance (Matthews Correlation Coefficient of 0.89). The proposed strategy could be implemented for the evaluation of chemicals in the hit/lead optimization stages of the drug discovery process, improve the selection of drug candidates as well as the development of safe pharmaceutical products.

Graphical abstracta) Enrichment plot of hERG and b) receiver operating characteristic (ROC) plots of hERG liability model built on the basis of the test set. Area under the curve (AUC) of 0.94 indicates the quality of this model.Figure optionsDownload full-size imageDownload as PowerPoint slide