SMILES parsing should be offered as a central module to all predictors. This avoids issues such as one model being unable to parse a SMILES that can be parsed by another model. Also, it is more efficient.
We should base the SMILES parsing on libRDChebi to ensure that we can parse the SMILES that ChEBI produces. For an example where standard SMILES parsing fails, see ChEB-AI/chebifier-web#24