Should MolToSmiles/Inchi respect enhanced stereo of singleton &1s?

[pechersky]

I'm asking about 1-atom STERAC groups, like the following:

  ChemDraw11132320552D

  0  0  0     0  0              0 V3000
M  V30 BEGIN CTAB
M  V30 COUNTS 4 3 0 0 0
M  V30 BEGIN ATOM
M  V30 1 C -0.714471 -0.618750 0.000000 0
M  V30 2 C -0.000000 -0.206250 0.000000 0
M  V30 3 Cl 0.714471 -0.618750 0.000000 0
M  V30 4 F -0.000000 0.618750 0.000000 0
M  V30 END ATOM
M  V30 BEGIN BOND
M  V30 1 1 2 1
M  V30 2 1 2 3
M  V30 3 1 2 4
M  V30 END BOND
M  V30 END CTAB
M  END
$$$$

  ChemDraw11132320552D

  0  0  0     0  0              0 V3000
M  V30 BEGIN CTAB
M  V30 COUNTS 4 3 0 0 0
M  V30 BEGIN ATOM
M  V30 1 C -0.714471 -0.618750 0.000000 0
M  V30 2 C -0.000000 -0.206250 0.000000 0
M  V30 3 Cl 0.714471 -0.618750 0.000000 0
M  V30 4 F -0.000000 0.618750 0.000000 0
M  V30 END ATOM
M  V30 BEGIN BOND
M  V30 1 1 2 1 CFG=1
M  V30 2 1 2 3
M  V30 3 1 2 4
M  V30 END BOND
M  V30 BEGIN COLLECTION
M  V30 MDLV30/STERAC1 ATOMS=(1 2)
M  V30 END COLLECTION
M  V30 END CTAB
M  END
$$$$

  ChemDraw11132320552D

  0  0  0     0  0              0 V3000
M  V30 BEGIN CTAB
M  V30 COUNTS 4 3 0 0 0
M  V30 BEGIN ATOM
M  V30 1 C -0.714471 -0.618750 0.000000 0
M  V30 2 C -0.000000 -0.206250 0.000000 0
M  V30 3 Cl 0.714471 -0.618750 0.000000 0
M  V30 4 F -0.000000 0.618750 0.000000 0
M  V30 END ATOM
M  V30 BEGIN BOND
M  V30 1 1 2 1
M  V30 2 1 2 3
M  V30 3 1 2 4 CFG=1
M  V30 END BOND
M  V30 BEGIN COLLECTION
M  V30 MDLV30/STERAC1 ATOMS=(1 2)
M  V30 END COLLECTION
M  V30 END CTAB
M  END
$$$$

Chem.MolToSmiles for the latter two shows a stereocenter. Chem.MolToInchi also shows stereoblocks. Is this expected/desired?

[bp-kelley]

Smiles has no knowledge of extended stereo groups, CXSMiles, however, does:

data = """"<above>"""
for m in Chem.ForwardSDMolSupplier(BytesIO(data.encode())):
    print(Chem.MolToCXSmiles(m))

results in:

CC(F)Cl |(-0.714471,-0.61875,;0,-0.20625,;0,0.61875,;0.714471,-0.61875,)|
C[C@H](F)Cl |(-0.714471,-0.61875,;0,-0.20625,;0,0.61875,;0.714471,-0.61875,),wU:1.0,&1:1|
C[C@H](F)Cl |(-0.714471,-0.61875,;0,-0.20625,;0,0.61875,;0.714471,-0.61875,),wU:1.1,&1:1|

[bp-kelley]

This is the ChemAxon extended smiles (btw):

https://docs.chemaxon.com/display/docs/chemaxon-extended-smiles-and-smarts-cxsmiles-and-cxsmarts.md

[pechersky]

Right. I understand that SMILES doesn't understand extended stereo groups. My question is regarding how much "canonicalization" I should be expecting that RDKit does before/during MolToSmiles when putting in a molecule with such &1s. It already does some type of canonicalization over atom order (and thus CW/CCW inference). And during MolToInchi, there is an internal tautomer canonicalization path.

[bp-kelley]

It really does sound like CXSmiles is what you want here, it does canonicalize the "&"s.

from rdkit.Chem.RegistrationHash import GetStereoTautomerHash

for i,m in enumerate(Chem.ForwardSDMolSupplier(BytesIO(data.encode()))):
    m.RemoveAllConformers() # remove coords
    print("smi", Chem.MolToSmiles(m))

for i,m in enumerate(Chem.ForwardSDMolSupplier(BytesIO(data.encode()))):
    m.RemoveAllConformers() # remove coords
    print("cxsmi", Chem.MolToCXSmiles(m))

Results in

0 smi  : CC(F)Cl
1 smi  : C[C@@H](F)Cl
2 smi  : C[C@@H](F)Cl

0 cxsmi: CC(F)Cl
1 cxsmi: C[C@H](F)Cl |&1:1|
2 cxsmi: C[C@H](F)Cl |&1:1|

[bp-kelley]

Note, you might want to remove coordinates when outputting cxsmiles for canonicalization purposes.

[kienerj]

Wouldn't the cleaner way be to use the appropriate flag, CX_ALL_BUT_COORDS?

Chem.MolToCXSmiles(m, flags=CXSmilesFields.CX_ALL_BUT_COORDS)

[bp-kelley]

Sometimes I can’t keep the whole API in my head :)

And yes, this is more clear.

[pechersky]

Do you think there is a way to get a "spirit of the law" way to say these 3 molecules represent the same mixture? One option could be comparing the sets of SMILES strings that the enumerator produces.

[bp-kelley]

I think that currently, generating the isomers is probably the best way, we do have. Registration Hash but it currently has different results for 0 and (1,2)

# enumerate isomers smiles into set()
... print(".".join(sorted(isomer_smi))

0 isomers C[C@@H](F)Cl.C[C@H](F)Cl
1 isomers C[C@@H](F)Cl.C[C@H](F)Cl
2 isomers C[C@@H](F)Cl.C[C@H](F)Cl

Currently we can't add 'missing' & stereo groups. That might be a nice feature, actually.

[pechersky]

This works for singleton &1, but doesn't work for singleton or1, which are supposed to indicate discernable molecules:

print(Chem.MolToCXSmiles(Chem.MolFromSmiles("C[C@H](F)Cl |o1:1|")))  # C[C@H](F)Cl |o1:1|
print(Chem.MolToCXSmiles(Chem.MolFromSmiles("C[C@@H](F)Cl |o1:1|")))  # C[C@H](F)Cl |o1:1|

for smiles in ["C[C@H](F)Cl |o1:1|", "C[C@@H](F)Cl |o1:1|"]:
  mol = Chem.MolFromSmiles(smiles)
  isomers = tuple(EnumerateStereoisomers(mol))
  print(".".join(sorted(Chem.MolToSmiles(x, isomericSmiles=True) for x in isomers)))

# C[C@@H](F)Cl.C[C@H](F)Cl
# C[C@@H](F)Cl.C[C@H](F)Cl

[bp-kelley]

I think the missing piece is the number of isomers “in the sample”.

&1 == two isomers
Or1 == one isomer out of 2

I don’t think we have a direct way of calculating this.

What, exactly, are you trying to do?

[bp-kelley]

However, to answer your question, SMILES was never intended to solve this particular problem that extended stereo chemistry solves, so it should not consider it in canonicalization. While inchi does normalize some tautomers, it also has the same issue regarding extended stereochemistry.

[bp-kelley]

Loosely I consider the smiles notation to indicate what a molecule IS and extended stereo chemistry to indicate what a molecule could be.

[pechersky]

I'm trying to understand which string representation of a molecule throws or keeps information in what way. @coleb suggested I contribute the following two functions:
Chem.EnhancedStereoToTraditionalStereo, and back again Chem.TraditionalStereoToEnhancedStereo that did the proper conversion of singleton &1 and abs tags

[pechersky]

One application for enhanced->traditional conversion is depiction. Depicting with hash+&1 can be somewhat difficult to spot, especially if you have two racemic stereocenters, then you have to notice that the &1 and &2 are referring to different numbers. Versus the same number -- very different mixture. Additionally, "canonicalizing" the depiction to a flat bond might be better than not canonicalizing at all (hash+&1 and wedge+&1 look different, but are really depicting the same mixture), or canonicalizing to a preferred lets say hash+&1.

[bp-kelley]

CXSmiles already canonicalizes (hash+&1 and wedge+&1)

[pechersky]

Yes, depiction code paths need to know, "ah I must do ROMol -> CXSmiles -> ROMol -> depict, not just ROMol -> depict". And OE toolkits don't canonicalize CXSMILES =C

[bp-kelley]

Would it help if these were options in the depictor system?

[bp-kelley]

I.e. Render assumed enhance stereo?