SMILES Pair Encoding (JCIM) first learns a vocabulary of high frequency SMILES substrings from a large chemical dataset (e.g., ChEMBL) and then tokenizes SMILES based on the learned vocabulary for ...