Using Chem.CanonSmiles in stk.Smiles() for better canonicalisation versus canonicalising the molecule.

[andrewtarzia]

I have had an example where the canonicalisation in stk.Smile().get_key(stk.Molecule) does not find equivalent molecules when they are built from different topology graphs in stk. But if I use Chem.CanonSmiles on the smiles from those molecules through rdkit, they are found to be matching.

Should we change algorithms? Or add a secondary algorithm? Or a warning?

Example of this implementation can be found in the new stko example: https://github.com/andrewtarzia/stko/blob/a57cec0ccac12670ad5b1a8209a783d23de64855/examples/intermediates_example.py

To implement the class: stko.UnreactedTopologyGraph, I needed to check for duplicate smiles in the get_reacted_smiles method, which would not work with the stk implementation of canonicalisation:

    def get_reacted_smiles(self, n: int | None = None) -> set[str]:
        """Yield constructed molecules with n reactions performed."""
        yielded_smiles = set()
        for const_mol in self.yield_constructed_molecules(n=n):
            distinct_molecules = self.separate_molecule(const_mol)
            for dmol, _ in distinct_molecules:
                smiles = Chem.CanonSmiles(
                    Chem.MolToSmiles(dmol.to_rdkit_mol())
                )

                if "." in smiles:
                    msg = "Found `.` in smiles."
                    raise RuntimeError(msg)

                if smiles in yielded_smiles:
                    continue
                yielded_smiles.add(smiles)

        return yielded_smiles

[lukasturcani]

link example for posterity?