--- license: openrail --- This dataset consists of Pubchem molecules downloaded from: https://ftp.ncbi.nlm.nih.gov/pubchem/Compound/CURRENT-Full/ There are in total ~85M compounds for training, with an additional ~10M held out for validation and testing.