random selfies to valid smiles

[DawnJYe]

Hi！Can this convert any combination of selfies into valid smiles? I converted random selfies into smiles, and then used rdkit's Chem.MolFromSmiles method to convert the smiles, but the result was empty. Does this mean that the smiles converted from selfies are invalid?
Thank you！

[MarioKrenn6240]

Hi @DawnJYe -- Try this code, hope it works for you. Yes arbitrary SELFIES can lead to a semantically and syntactically valid molecule.

[DawnJYe]

valid
Thank you for your response. If I create a selfies dictionary using my own smiles, will any combination still be valid? I tried using a custom dictionary to form selfies and encountered an error when running it through the selfies.decoder. Do I have to use the get_semantic_robust_alphabet() function instead?

[MarioKrenn6240]

They should still be valid. If not, i guess there is a problem somewhere. Can you share cases which fail? (Note: We regard an empty string as valid, maybe thats the difference to RDKit's output?)

[DawnJYe]

Do you mean that after passing selfies through a decoder, the resulting smiles may be an empty string? If so, then I think I'm encountering that situation. The error message I'm getting is “rdkit.Chem.rdmolfiles.MolToSmiles(NoneType)”.So is this situation considered valid in selfies and it's just that RDKit doesn't accept empty strings?

[DawnJYe]

They should still be valid. If not, i guess there is a problem somewhere. Can you share cases which fail? (Note: We regard an empty string as valid, maybe thats the difference to RDKit's output?)

Sorry for interrupting again. I am facing a situation where selfies is represented as '[C][C][Sn][Branch1][Ring1][C][C][Branch1][#Branch2][C][=C][C][=C][C][=C][Ring1][=Branch1][C][Ring1][#Branch1][C][Ring1][=Branch1][C]', but after passing it through selfies.decoder, the generated SMILES is 'C1C[Sn]1(CC)(C2=CC3=CC=C2C)C3C'. However, it seems to be invalid for RDKit, as Chem.MolFromSmiles returns None when using RDKit. Could you please tell me what the reason for this is?

[robpollice]

Are you using the default SELFIES constraints? If yes, the problem is that in the default constraints, Sn does not have specific constraints. That means that the default behavior is allowing for 8 valences, which is not meaningful for Sn. What this means is that you need to create your custom SELFIES constraints that constrain Sn to a meaningful number of valences, say 4 for instance, and then use SELFIES with these custom constraints.

[MarioKrenn6240]

Small addition, info to customizing selfies is here, and in chapter 4.3. "Customization functions" of our software-paper.