Insights for human ether-a-go-go-related gene potassium channel inhibition using recursive partitioning and kohonen and sammon mapping techniques

The human ether-a-go-go related gene (hERG) can be inhibited by marketed drugs, and this inhibition may lead to QT prolongation and possibly fatal cardiac arrhythmia. We have collated literature data for 99 diverse hERG inhibitors to generate Kohonen maps, Sammon maps, and recursive partitioning models. Our aim was to investigate whether these computational models could be used either individually or together in a consensus approach to predict the binding of a prospectively selected test set of 35 diverse molecules and at the same time to offer further insights into hERG inhibition. The recursive partitioning model provided a quantitative prediction, which was markedly improved when Tanimoto similarity was included as a filter to remove molecules from the test set that were too dissimilar to the training set (r(2) = 0.83, Spearman rho 0.75, p = 0.0003 for the 18 remaining molecules, > 0.77 similarity). This model was also used to screen and prioritize a database of drugs, recovering several hERG inhibitors not used in model building. The mapping approaches used molecular descriptors required for hERG inhibition that were not reported previously and in particular highlighted the importance of molecular shape. The Sammon map model provided the best qualitative classification of the test set (95% correct) compared with the Kohonen map model (81% correct), and this result was also superior to the consensus approach. This study illustrates that patch clamping data from various literature sources can be combined to generate valid models of hERG inhibition for prospective predictions.