MolecularAI / aizynthtrain

Tools to train synthesis prediction models
Apache License 2.0
21 stars 7 forks source link

filtering out the reactions which the number of product is greater than 1 #5

Closed yangxfei closed 10 months ago

yangxfei commented 1 year ago

hello, folks: when cleaning reaction smiles, could you help answer why need to remove reactions whose products are greater than 1? could you explain more about this limitation? My understanding is that the input of the modeling is based on mol's fingerprint, it can't indicate the mutiple product's fingerprint as input.

Thanks.

Philip Yang

I note below

Removing reactions based on these filters

yangxfei commented 1 year ago

the reason that I have this question is it is popular that many reactions lead to mutiple products, other side product may include h2o and co2 etc. For such reactions, we may want to keep these reactions during trainning.

Thanks.

SGenheden commented 11 months ago

The reason for keeping 1 product is that these models are trained for retrosynthesis where you go from 1 product to 1 or more reactants. These reactions are never balanced, so by/side-products like H2O or CO2 are not included.

SGenheden commented 10 months ago

Closing due to inactivity