A Virtual Screening Method for Prediction of the hERG Potassium Channel Liability of Compound Libraries

A computer‐based method has been developed for prediction of the hERG (human ether‐à‐go‐go related gene) K+‐channel affinity of low molecular weight compounds. hERG channel blockage is a major concern in drug design, as such blocking agents can cause sudden cardiac death. Various techniques were applied to finding appropriate molecular descriptors for modeling structure–activity relationships: substructure analysis, self‐organizing maps (SOM), principal component analysis (PCA), partial least squares fitting (PLS), and supervised neural networks. The most accurate prediction system was based on an artificial neural network. In a validation study, 93 % of the nonblocking agents and 71 % of the hERG channel blockers were correctly classified. This virtual screening method can be used for general compound‐library shaping and combinatorial library design.