Introduction

Modeling hERG inhibition has significantly gained popularity since 2005, when the FDA recognized the correlation between hERG inhibition and a prolonged QT interval by issuing guidance for the evaluation of new non-antiarrythmic drugs against the hERG channel. Long QT syndrome or LQTS is a risk factor  for ventricular tachyarrhythmias and sudden death.

Here we present the evolution of our hERG inhibition model in consecutive releases of ADMET Predictor™. Examples detailing the impact of new and evolving descriptors on the “TOX_hERG” model’s applicability domain and performance on internal and external data are provided. Focus is given to a particularly interesting case where an earlier release of ADMET Predictor outperformed its successor on a client’s proprietary data. Finally, we discuss how we are improving model selection criteria through the use of descriptor sensitivity analysis with artificial neural network ensembles in combination with a better understanding of the model’s applicability domain, based on the World Drug Index

242nd ACS National Meeting, August 28-September 1, 2011, Denver CO

By Adam C. Lee, Grazyna Fraczkiewicz, Robert Fraczkiewicz, Robert Clark, Walter S. Woltosz