Khac - Minh Thai

Title of the Doctoral Thesis: In silico models for prediction of hERG potassium channel blockers.

Publishing year: 2008

Tags: hERG / QSAR / classification / binary QSAR / counter-propagation / neural network / SIBAR / docking / propafenone / drug trapping / GH score


Inhibition of human ether-a-go-go-related-gene (hERG) channels prolongs the ventricular action potential with the risk of torsade de pointes arrhythmia that may result in sudden cardiac death. Therefore, computational approaches for classification and prediction of hERG affinity in an early phase of the drug discovery and development process are of increasing interest. Both structure-based and ligand-based approaches have been undertaken to shed more light on the molecular basis of drug-channel interaction as well as to predict hERG affinity and to classify hERG inhibitors. In this study, an in silico system for classification and prediction hERG blockers is reported which combines ligand-based (binary QSAR and counter-propagation neural networks (CPG-NNs)) and structure-based approaches. Binary QSAR models with different threshold values were generated using several sets of descriptors including P_VSA descriptors, similarity based SIBAR descriptors and a set of 2D descriptors identified out of a large set via a feature selection algorithm. The power for classification of hERG blockers by binary QSAR is high (0.82-0.88) and meets or even outperforms computational models published so far. Due to the possibility to define more than two classes, the CPG-NN with a 3-dimensional output layer provides possible strategies for improving the performance of predicting and classifying compounds which belong to the middle activity class (hERG IC50 =1-10 μM). We also investigated the performance of similarity-based descriptors (SIBAR) and evaluated the influence of the descriptors used for calculating the SIBAR values and the composition of the reference set. Best performance was achieved when both the descriptors and the reference set are related to hERG channel activities. With respect to the drug trapping property of the hERG channel, a systematic analysis of ‘use-dependent’ block and recovery of hERG channel were applied for a small set of propafenone derivatives. Docking runs provided insights into ligand binding modes at the open and closed states of the hERG channel which might explain differences in hERG potency and drug trapping. Recent studies on hERG functions showed that the hERG K+ channel is not only an antitarget but might also be a drug target, especially in oncology and cardiology. Hence, these predictive systems might be applicable both for the design of new hERG channel blockers as well as for early detection of possible undesired hERG activities.