David Ryan Koes · Dec 15

I'm not too worried about the learned descriptor results as they were specifically trained to co-locate structurally dissimilar, but biochemically similar, pockets. But this does point out that what a neural network sees as similar is not necessarily what a human would label as similar.