Selfies alphabet token changes

[whitead]

sf.encoder('C#[S-]')

outputs

'[C][#S-expl]'

which uses a token [#S-expl] that does not appear on get_semantic_robust_alphabet. The closest token I see is [#S+1expl].

This arises by running the following code:

start_selfies = '[C][#S-1expl]' 
start_smiles = sf.decoder(start_selfies)
end_smiles = rdkit.Chem.MolToSmiles(rdkit.Chem.MolFromSmiles(start_smiles))
print('smiles change', start_smiles, end_smiles)
end_selfies = sf.encoder(end_smiles)
print('selfies change', start_selfies, end_selfies)

smiles change C#[S-1] C#[S-]
selfies change [C][#S-1expl] [C][#S-expl]

Is this intended behavior? I'm having this happen while running STONED and it's changing my alphabet. Any ideas would be appreciated!

[MarioKrenn6240]

Thanks for this comment. We will constrain the decoder to output only one version, [X-1expl] in the next version (after the workshop).

[alstonlo]

Hi @whitead,

In selfies 2.0.0, we have constrained selfies.encoder to one unique representation for every atom, which also matches the symbols returned by selfies.get_semantic_robust_alphabet. This should prevent alphabet changes! For example,

print(sf.encoder("C#[S-1]"))  # [C][#S-1]
print(sf.encoder("C#[S-]"))   # [C][#S-1]
print(sf.encoder("[CH]#C"))   # [CH1][#C]
print(sf.encoder("[CH1]#C"))  # [CH1][#C]

Thanks for the suggestion!

[whitead]

Sounds good to me. Thanks!