Performance of chemical structure string representations for chemical image recognition using transformers dataset

The datasets contain string representations used for DECIMER short communication paper.

ChEMBL dataset:

Train and test datasets downloaded from ChEMBL and curated. Contains data with and without stereochemistry. Separated as Canonical and Isomeric.

String representations contain SMILES, DeepSMILES, SELFIES and InChIs.

• Train dataset: 1.5 Mio molecules
• Test dataset: ~100K molecules

Pubchem dataset:

Train and test datasets downloaded from PubChem and curated. Contains data with and without stereochemistry. Separated as Canonical and Isomeric.

String representations contain SMILES, DeepSMILES and SELFIES.

• Train dataset: 3 Mio molecules
• Test dataset: 250K molecules