Nonlinear classification of hERG channel inhibitory activity by unsupervised classification method

The side effects that occur in the central nervous system and circulatory system due to medicines are expected to be prevented by research and development. However, many of the compounds in medicines have the possibility of causing arrhythmia, and methods developed to detect this problem at the early stage of drug development are not always successful. In the present study, we classified drug compounds according to their activity using only structural information. To classify compounds, we used a self-organizing map (SOM), which is a nonlinear unsupervised classification method. We first analyzed a small-scale dataset, and an excellent classification result was obtained. We then applied our method to a large-scale dataset containing numerous inert compounds and were again able to classify the compounds according to their activity. Both classifications showed some compound activity, although a few differences between the two SOM maps were seen.