At the lead generation stage for drug discovery, the efficient finding of hit compounds is a key to success. In this paper, we have applied the active learning method as an effective compound screening method and have shown its effectiveness by both computer simulations using known chemical data and actual wet experiments. With regard to the computer simulations, it is shown that one fifth screening is enough for finding ninety percent of all hit compounds. Our method could lessen eighty percent of actual wet experiments. We have performed actual binding experiments, and have also shown that active learning could find almost all 'hits' with a reduced number of actual binding experiments.
展开▼
