The blockage of the hERG potassium channel by a wide number of diverse compounds has become a major pharmacological safety concern as it can lead to sudden cardiac death. In silico models can be potent tools to screen out potential hERG blockers as early as possible during the drug-discovery process. In this study, predictive models developed using the recursive partitioning method and created using diverse datasets from 203 molecules tested on the hERG channel are described. The first model was built with hERG compounds grouped into two classes, with a separation limit set at an IC50 value of 1 microm, and reaches an overall accuracy of 81%. The misclassification of molecules having a range of activity between 1 and 10 microM led to the generation of a tri-class model able to correctly classify high, moderate, and weak hERG blockers with an overall accuracy of 90%. Another model, constructed with the high and weak hERG-blocker categories, successfully increases the accuracy to 96%. The results reported herein indicate that a combination of precise, knowledge management resources and powerful modeling tools are invaluable to assessing potential cardiotoxic side effects related to hERG blockage.

Based on a basic principle assuming that similar molecules share similar target activity, new potential targets and, therefore, opportunities of potential new indications have been identified and discussed.

No abstract

No abstract

KEy worDsion channel, biological space, drug design, chemical diversity, Structure-activity relationships (SAR) AcKnowlEDgEmEntsThe authors express their appreciation to ChemAxon team for providing JChem tools, and their helpful support. We thank Sophie Ollivier and the knowledge management team as well as Dominique Neaud and the IT team for their valuable help during the preparation of this work. We gratefully thank Mary Donlan for her comments and for assisting us with the proofreading of the manuscript. Emmanuel Bourinet is supported by operating grants from the Agence Nationale de la Recherche (ANR-05-NEUR-031-01), the ARC-INCa-2006, the Institut UPSA de la Douleur, the Association Française contre les Myopathies (AFM), and the Fédération pour la Recherche sur le Cerveau (FRC, équipe-ment 2006). ABstrActThe aim of the present work is to assess the chemical and biological diversity of ligands reported in scientific articles or patents to be active against ion channels targets. A specific query of the AurSCOPE Ion Channel knowledge database was constructed to retrieve a set of the most active non-peptide ligands tested in binding or electrophysiology experiments against all ion channel families. A biological activity threshold cutoff expressed by K i , IC 50 , or EC 50 was set to 300 nM. This activity cutoff was selected such that we would retrieve a set of compounds, which contain the most active ligands for all target families, but is a reasonable number to analyze. To encode the chemical space for the entire active dataset (9897 molecules), ChemAxon's chemical fingerprints were computed and optimized and then employed to cluster the dataset at a variety of different similarity thresholds. Concurrently, the exploration of the biological space was performed by associating with each chemical cluster the corresponding target or target family. Tri-dimensional visualization of different voltage-and ligand-gated ion channel families projected into the active chemical space was obtained after a principal components analysis performed using selected molecular descriptors. The findings presented herein give a global picture of the realm of ion channels active ligands and link the knowledge on chemical structures with their respective biological activities.