`MolToSmiles` fails with `rootedAtAtom` for disconnected molecules (need for smiles augmentation)

[vidvath7]

rdkit MolToSmiles() gives RuntimeError for some molecules (Eg: "[Cl-].[H]N+([H])[H]" ), with configuration of rootedAtAtom>0.

[aditya0by0]

I'm facing the same error for issue #113 in PR #115 , and the below might the reason

Description:
When performing SMILES augmentation using RDKit’s Chem.MolToSmiles with the rootedAtAtom argument, the function raises a RuntimeError: Range Error idx for molecules containing disconnected fragments (e.g., [F-].[Na+]).

Steps to Reproduce:

from rdkit import Chem

mol = Chem.MolFromSmiles("[F-].[Na+]")
Chem.MolToSmiles(mol, rootedAtAtom=1, doRandom=True)

Observed Behavior:

Exception has occurred: RuntimeError
Range Error
	idx
	Violation occurred on line 209 in file Code\GraphMol\ROMol.cpp
	Failed Expression: 1 < 1
	RDKIT: 2024.09.6
	BOOST: 1_85

Cause:
RDKit’s rootedAtAtom assumes the atom index belongs to a single connected component. For disconnected molecules (multiple fragments), this internal assumption fails and results in an out-of-bounds fragment index.

To elaborate, MolToSmiles’ rootedAtAtom parameter expects the index of atom in a connected fragment — but the molecule [F-].[Na+] is two disconnected fragments. When you pass rootedAtAtom=1 (sodium atom), RDKit internally tries to traverse a single fragment, but the fragment index calculation fails due to it being a disconnected molecule.

@sfluegel05, Please let me know if you have any suggestions for augmenting this type of smiles which are present in our processed dataset.

[sfluegel05]

you are right. For some reason the rootedAtAtom argument does not accept the atom id, but only takes the first fragment into account. I guess we should treat each fragment separately as well and then recombine them afterwards

mol = Chem.MolFromSmiles("[F-].[Na+]")
fragments = Chem.GetMolFrags(mol, asMols=True)
for i, fragment in enumerate(fragments):
    print(f"Fragment {i}: {Chem.MolToSmiles(fragment)}")
    # randomize SMILES for this fragment
# stitch together fragment-SMILES (in random order)