Fig. 8

Kekulization differences: Reaction a extracted from the USPTO dataset is parsed by CDK and its SMIRKS is recreated after removing the agents (Reaction b). The only difference in the regenerated SMIRKS is the atom order. Despite all other characteristics remaining unchanged (atom-atom mapping, aromaticity, implicit hydrogens, etc.), the resulting SMIRKS leads to a different kekulization (kekulization of the ring containing atoms 10–15 in the reactants)