how to handle tautomers? about molpal HOT 2 CLOSED

Rather than having multiple tautomers as distinct library members, I imagine the best solution would be to have a canonicalization pipeline that tries to standardize representations so this isn't an issue, e.g., using RDKit's standardization with MolVS or roundtripping to/from InChI (if appropriate)

from molpal.

thanks for your reply, this is really helpful.

I am not an expert on machine learning, naively I thought it is a good thing that providing more information(tautomers) for training the surrogate model.

so this is just not the case, I think. Thank you!

from molpal.