what data used for prior

[rneeser]

Hi! Thanks for this awesome collection! I was wondering what dataset exactly was used to train the model random.prior.new. I assume this is the version using data from ChEMBL with randomized SMILES? What additional filters where there applied in terms of:

• element types
• number of heavy atoms
• max number of e.g. rings with or without heteroatoms
• deduplication based on stereochemistry

Thanks!