SetRemoveSp3Stereo(False) does not keep isotope (deuterium) induced stereochemistry

[pechersky]

Describe the bug
Canonicalizing a molecule with implicit-ized isotopic Hs removes the stereo that those deuteriums might induce. The implicitization of these Hs is necessary to overcome the issue described in #5340. Perhaps @ptosco or @kienerj have comments.

To Reproduce

from rdkit import Chem
from rdkit.Chem.MolStandardize import rdMolStandardize

smiles = [
    # these are fine
    r"FC(C)C",
    r"F[C@H](C)CC",
    r"F[C@@H](C)CC",
    r"FCC",
    # and these lose stereo
    r"F[C@H]([2H])C",
    r"F[C@@H]([2H])C",
]

for smi in smiles:
    mol = Chem.MolFromSmiles(smi)
    had_at_sign = "@" in smi
    params = Chem.RemoveHsParameters()
    params.removeAndTrackIsotopes = True
    mol = Chem.RemoveHs(mol, params)
    enumerator = rdMolStandardize.TautomerEnumerator()
    enumerator.SetRemoveSp3Stereo(False)
    mol = Chem.AddHs(enumerator.Canonicalize(mol), explicitOnly=True)
    print(("@" in Chem.MolToSmiles(mol)) == had_at_sign)

Expected behavior
I expect the sp3 stereo of the carbons (due to D stereo) to remain. This means all the output of the above script should be True.

Screenshots

Configuration (please complete the following information):

• RDKit version: 2024.3.1
• OS: [e.g. Ubuntu 20.04]: ubuntu 20.04
• Python version (if relevant): py3.11
• Are you using conda? no
• If you are using conda, which channel did you install the rdkit from? self-built(rdkit-pypi style)
• If you are not using conda: how did you install the RDKit? pip install from wheel

Additional context
Add any other context about the problem here.

[greglandrum]

@pechersky This isn't really a problem with the tautomer enumeration: it is not possible to preserve stereochemistry when you remove one of the atoms which determine the stereochemistry.
Here's a quick demo of what happens:

In [5]: m = Chem.MolFromSmiles('F[C@@H]([2H])C')

In [6]: params = Chem.RemoveHsParameters()

In [7]: params.removeAndTrackIsotopes = True

In [8]: nm = Chem.RemoveHs(m,params)

In [9]: Chem.MolToSmiles(nm)
Out[9]: 'CCF'

In [11]: Chem.AssignStereochemistry(nm,force=True,cleanIt=True)

In [12]: nm.Debug()
Atoms:
	0 9 F chg: 0  deg: 1 exp: 1 imp: 0 hyb: 4 arom?: 0 chi: 0
	1 6 C chg: 0  deg: 2 exp: 4 imp: 0 hyb: 4 arom?: 0 chi: 0
	2 6 C chg: 0  deg: 1 exp: 1 imp: 3 hyb: 4 arom?: 0 chi: 0
Bonds:
	0 0->1 order: 1 conj?: 0 aromatic?: 0
	1 1->2 order: 1 conj?: 0 aromatic?: 0

As an aside: this example, where you have a chiral center because a C has one H and one D attached, strikes me as highly artifiical. Do you have a real-world example of this?

[pechersky]

Without removing (and tracking) these deuteriums, then the tautomer enumeration does not shift around the explicit Hs (or Ds), thus creating, in some instances, a bad valence "canonical tautomer". This is what came up in the issue referenced above.

We're working on shoring up our canonicalization-and-comparison tools, and part of them is considering isotopically induced stereochemistry (heavy atom or not). Our current approach treats all elements equally -- but this ran aground of the molecule cleanup we do before tautomer canonicalization in this D/T case.

One option we might take is "we're never going to have isotopically labeled Hs on tautomeric centers so it's fine to clear them away".

[greglandrum]

I think there's a case to be made for an enhancement request to allow the tautomer enumeration to work on molecules with Hs in the graph. That's a non-trivial change, but it is theoretically doable. But that's not this.

I wonder if there's any real-world situation (i.e. something that has actually been done in the lab and not just constructed as a what if? example) where a D in place of an H makes a molecule chiral in a way that matters

[pechersky]

Real world situations:
Chemical-shift isotope effect (NMR): Lambert, Joseph B., and Linda G. Greifenstein. "Stereochemical dependence of the chemical-shift isotope effect." Journal of the American Chemical Society 95.18 (1973): 6150-6152.
Chiral catalyst (chirality due to H-D chirality in the catalyst): Sato, Itaru, et al. "Highly enantioselective synthesis induced by chiral primary alcohols due to deuterium substitution." Journal of the American Chemical Society 122.47 (2000): 11739-11740.
Diastereomerically pure bicycles due to trans or cis H-D:
Saito, Fumito, et al. "Breaking the Symmetry of a Meso Compound by Isotopic Substitution: Synthesis and Stereochemical Assignment of Monodeuterated cis-Perhydroazulene." Organic Letters 23.1 (2020): 113-117.

The chiral catalyst paper is particularly relevant because the catalysts are c1ccccc1[C@H]([2H])O, which, if tautomer-canonicalization was applied to, might have issues in the keto-enol reactions (?)

I stereoenumerate molecules and then pose them into 3D in pockets. Assuming that my stereoenumeration code is cognizant to generate 2 enantiomers of something like FC[2H] and 3D conformer generation respects the stereochemistry, differentiating the two allows for downstream analysis for something like positioning of the D in an enzyme's active site and whether that would affect the enzyme's catalytic cycle.