Inconsistent use of mapped SMILES breaks internal consistency assumptions #206

Activity

1. mattwthompson commented on Oct 20, 2026

   

   mattwthompson

   on Oct 20, 2026

   MemberAuthor

   More actions

   This doesn't trigger when calling _gather_from_reference on identical force fields

   I'm able to pinpoint it to chemically equivalent SMILES being written slightly differently at some point, but the provenance of how these two strings got onto my local files is lost to the sands of time. I don't think I can easily pin it to OEChem vs. RDKit differences:

    ipdb> {key.id for key in ref.parameter_keys if key.associated_handler == "NAGLChargesHandler" and len(key.id) == 73}
   {'[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'}
   ipdb> {key.id for key in potential.parameter_keys if key.associated_handler == "NAGLChargesHandler" and len(key.id) == 73}
   {'[H:11][c:5]1[c:4]([c:3]([c:2]([c:1]([c:6]1[H:12])[H:7])[H:8])[H:9])[H:10]'}
   ipdb> from openff.toolkit.utils import *
   ipdb> openff.toolkit.Molecule.from_smiles("c1ccccc1").to_smiles(mapped=True, toolkit_registry=OpenEyeToolkitWrapper())
   '[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'
   ipdb> openff.toolkit.Molecule.from_smiles("c1ccccc1").to_smiles(mapped=True, toolkit_registry=RDKitToolkitWrapper())
   '[c:1]1([H:7])[c:2]([H:8])[c:3]([H:9])[c:4]([H:10])[c:5]([H:11])[c:6]1[H:12]'
2. mattwthompson commented on Oct 20, 2026

   

   mattwthompson

   on Oct 20, 2026

   MemberAuthor

   More actions

   Okay so it's probably not OECHem vs. RDKit differences or Isomeric vs. non-isomeric SMILES (via the toolkit), but I do notice that the map_smiles function in this project produces a different string:

   >>> smiles = 'c1ccccc1'
   >>> mapped_smiles = '[H:11][c:5]1[c:4]([c:3]([c:2]([c:1]([c:6]1[H:12])[H:7])[H:8])[H:9])[H:10]'
   >>> import openff.toolkit
   >>> from openff.toolkit.utils import OpenEyeToolkitWrapper, RDKitToolkitWrapper
   >>> openff.toolkit.Molecule.from_mapped_smiles(mapped_smiles).to_smiles(mapped=True, isomeric=True, explicit_hydrogens=True, toolkit_registry=OpenEyeToolkitWrapper())
   '[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'
   >>> openff.toolkit.Molecule.from_mapped_smiles(mapped_smiles).to_smiles(mapped=True, isomeric=False, explicit_hydrogens=True, toolkit_registry=OpenEyeToolkitWrapper())
   '[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'
   >>> openff.toolkit.Molecule.from_mapped_smiles(mapped_smiles).to_smiles(mapped=True, isomeric=True, explicit_hydrogens=True, toolkit_registry=RDKitToolkitWrapper())
   '[c:1]1([H:7])[c:2]([H:8])[c:3]([H:9])[c:4]([H:10])[c:5]([H:11])[c:6]1[H:12]'
   >>> from tyff.utils.molecule import map_smiles
   >>> map_smiles(smiles)
   '[c:1]1([H:7])[c:2]([H:8])[c:3]([H:9])[c:4]([H:10])[c:5]([H:11])[c:6]1[H:12]'
   >>> mapped_smiles
   '[H:11][c:5]1[c:4]([c:3]([c:2]([c:1]([c:6]1[H:12])[H:7])[H:8])[H:9])[H:10]'

   This difference is relevant because openff.interchange.models.PotentialKey has its .id attribute set (as a simple Python string) by a separate Molecule.to_smiles call. That should all be self-consistent, but we cannot assume that the result of tyff.utils.molecule.map_smiles (or other to_smiles calls¹) will produce identical strings:

   >>> mapped_smiles
   '[H:11][c:5]1[c:4]([c:3]([c:2]([c:1]([c:6]1[H:12])[H:7])[H:8])[H:9])[H:10]'
   >>> {key.id for key in openff.toolkit.ForceField("openff-2.3.0.offxml").create_interchange(Molecule.from_mapped_smiles(mapped_smiles).to_topology())['Electrostatics'].potentials}
   {'[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'}
   >>> Molecule.from_mapped_smiles(mapped_smiles).to_smiles(mapped=True)
   '[H:9][c:3]1[c:2]([c:1]([c:6]([c:5]([c:4]1[H:10])[H:11])[H:12])[H:7])[H:8]'

   It'll take me a second to find where this mismatch happens, but I think it's likely to be somewhere inside of tyff

   Footnotes

   1. Molecule.from_mapped_smiles(...).to_smiles(mapped=True) returning something different than the input took me by surprise before thinking about it - I'm pretty sure that strings being character-for-character equivalent through the toolkit isn't a reasonable expectation ↩
3. mattwthompson

   changed the title [-]`gather_from_reference` chokes with `NAGLChargesHandler`[/-] [+]Inconsistent use of mapped SMILES breaks internal consistency assumptions[/+] on Oct 20, 2026
4. mattwthompson commented on Oct 20, 2026

   

   mattwthompson

   on Oct 20, 2026

   MemberAuthor

   More actions

   I fixed this somewhere in #190
5. mattwthompson

   closed this as completedon Oct 20, 2026