Details of the calculation

This server simulates the Action Potential (AP) modification by drug-induced blockade of the potassium channel hERG (Kv11.1) and the potassium channel KCNQ1 (Kv7.1). The prediction is carried out by a multiscale which works at two levels (as shown in the figure below): first, at the molecular level, it predicts the binding affinity of the drug to hERG and KCNQ1 and then, the predicted binding affinities were used as input for a cell simulation of the AP. Compounds with predicted high affinity for both channels can produce severe modification of the AP and have a high potential cardiotoxicity.

scheme LQT computation

(a) Conversion of formats

The OpenBabel programĀ  [1] is used to convert the smiles format of the molecule into the SDFile (sdf) format for the next calculations.

(b) Assignation of the protonation state

The correct protonation state at pH=7.4 is predicted and assigned by means of MoKa [2].

(c) Conversion to 3D structure

A reasonable 3D structure representing the extended conformation of the molecule is assigned by using CORINA [3].

(d) Molecular Docking

The compounds are docked into the binding site of the receptors using program GOLD [4], first into the open state of hERG described in Farid et al. [5], and then into the open state of KCNQ1 described in Lerche et al. [6]. The binding site is supposed to be in the S6 helical domains of the channels. Three genetic algorithm runs are submitted and the two best scored docking poses, as indicated by the ASP scoring function, are selected for the last steps. Note that due to the genetic algorithm used in GOLD, consecutive runs of the same target molecule could not reach the same docking solution.

(e) pIC50 prediction

The best docking solution obtained for each channel is imported to Pentacle [7], an advanced 3D QSAR tool based on the GRIND-2 alignment-free descriptors [8,9].

Pentacle can be found as part of the VPH ToolKit.

Currently, a predictive model compiling more than 350 hERG blockers and 150 KCNQ1 blockers is used as the training set. The experimental data used for the training set were mainly obtained from public patch-clamp assays, expressing hERG mainly in CHO and HEK cells. The submitted molecule is considered as the external set and its binding affinity is predicted.

The final results are expressed as the pIC50 of the target molecule blocking the hERG and KCNQ1 channels. The standard deviation of the final results can be considered of about 0.9 log units. Compound with predicted pIC50 > 5 can be considered as a potential potassium channel blocker.

We expect to improve the accuracy of the IC50 calculation using the results of currently running molecular dynamics simulations. Visit the GPUGRID website, ‘volunteer computing for biomolecules’.

(f) Action Potential calculation

The predicted results at molecular level are incorporated into an electrophysiological model of guinea pig cell by using the Hodgkin-Huxley formalism [10]. This server provides the ventricular Action Potential simulation in absence and presence (100 nM) of the target compound. A severe modification of the regular Action Potential could be associated to cardiotoxicity.

Two example models of dofetlide hERG blockade (in cellML language) are provided here for reference. Download the Endo.cellml and the Mcell.cellml models.


This server is available for academic purposes only. The Computer-Assisted Drug Design Lab declines any responsibility for correctness, completeness or quality of the predictions provided.


[1] OpenBabel,

[2] Milleti, F. et al. New and original pKa prediction method using grid molecular interaction fields. J Chem Inf Model 47, 2172-2181 (2007).

[3] Gasteiger, J, Rudolph, C and Sadowski, J. Automatic generation of 3D-atomic coordinates for organic molecules. Tetrahedron Comput Methodol 3, 537-547 (1990).

[4] Verdonk, M.L. et al. Improved protein-ligand docking using GOLD. Proteins 52, 609-623 (2003).

[5] Farid, R. et al. New insights about hERG blockade obtained from protein modelling, potential energy mapping, and docking studies. Bioorg Med Chem 14, 3160-3173 (2006).

[6] Lerche, C. et al. Chromanol 293B binding in KCNQ1 (Kv7.1) channels involves electrostatic interactions with a potassium ion in the selective filter. Mol Pharmacol 71, 1503-1511 (2007).

[7] Pentacle, Molecular Discovery Ltd.

[8] Pastor, M. et al. Grid-INdependent Descriptors (GRIND): a novel class of alignment-independent three dimensional molecular descriptors. J Med Chem 43, 3233-3243 (2000).

[9] Duran, A., Martinez, G.C. and Pastor, M. Development and validation of AMANDA, a new algorithm for selecting highly relevant regions in Molecular Interaction Fields. J Chem Inf Model 48, 1813-1823 (2008).

[10] Luo, H., Rudy, Y. A dynamic model of the cardiac ventricular action potential. I. Simulations of ionic currents and concentration changes. Circ Res 74, 1071-1096 (1994).