Bridgehead nitrogen stereo flipped on hydrogen removal

[mcs07]

Describe the bug

I noticed an interesting stereo-related issue when neutralizing a protonated atom, specifically when the atom continues to be a chiral center after the hydrogen is removed and the charge reduced.

To Reproduce

Taking quinine as an example molecule:

>>> from rdkit import Chem
>>>
>>> mol1 = Chem.MolFromSmiles("C=C[C@H]1C[N@@]2CC[C@H]1CC2[C@H](O)c1ccnc2ccc(OC)cc12")
>>> mol2 = Chem.MolFromSmiles("C=C[C@H]1C[N@@H+]2CC[C@H]1CC2[C@H]([O-])c1ccnc2ccc(OC)cc12")

mol1 is the neutral molecule and mol2 is a different protomer where the bridgehead nitrogen has a positive charge and an additional hydrogen. If we remove the hydrogen and reduce the formal charge, [N@@H+] has changed to [N@] in the resulting SMILES.

>>> mol2_n = mol2.GetAtomWithIdx(4)
>>> print(f"Element: {mol2_n.GetSymbol()} Charge: {mol2_n.GetFormalCharge():+} Hydrogens: {mol2_n.GetTotalNumHs()} ChiralTag: {mol2_n.GetChiralTag()}")
Element: N Charge: +1 Hydrogens: 1 ChiralTag: CHI_TETRAHEDRAL_CCW
>>> mol2_n.SetNumExplicitHs(0)
>>> mol2_n.SetNoImplicit(True)
>>> mol2_n.SetFormalCharge(0)
>>> Chem.SanitizeMol(mol2)
rdkit.Chem.rdmolops.SanitizeFlags.SANITIZE_NONE
>>> print(f"Element: {mol2_n.GetSymbol()} Charge: {mol2_n.GetFormalCharge():+} Hydrogens: {mol2_n.GetTotalNumHs()} ChiralTag: {mol2_n.GetChiralTag()}")
Element: N Charge: +0 Hydrogens: 0 ChiralTag: CHI_TETRAHEDRAL_CCW
>>> Chem.MolToSmiles(mol2)
'C=C[C@H]1C[N@]2CC[C@H]1CC2[C@H]([O-])c1ccnc2ccc(OC)cc12'

In this specific case, I know I can "fix" this by manually flipping CHI_TETRAHEDRAL_CCW to CHI_TETRAHEDRAL_CW, but I don't understand enough about how the CW/CCW accounting works for trivalent bridgehead nitrogens (for either the RDKit ChiralTag or the SMILES @/@@) to know if this is generally what should always be done.

>>> mol2_n.SetChiralTag(Chem.CHI_TETRAHEDRAL_CW)
>>> Chem.MolToSmiles(mol2)
'C=C[C@H]1C[N@@]2CC[C@H]1CC2[C@H]([O-])c1ccnc2ccc(OC)cc12'

I'm opening this as a bug report issue rather than a discussion because I think this situation may not be handled correctly internally to rdMolHash HetAtomTautomer/HetAtomProtomer and within the rdMolStandardize Uncharger:

>>> from rdkit.Chem import rdMolHash
>>> from rdkit.Chem.MolStandardize import rdMolStandardize
>>>
>>> rdMolHash.MolHash(mol1, rdMolHash.HashFunction.HetAtomTautomer)
'C=C[C@H]1C[N@@]2CC[C@H]1CC2[C@H]([O])[C]1[CH][CH][N][C]2[CH][CH][C](OC)[CH][C]21_1_0'
>>> rdMolHash.MolHash(mol2, rdMolHash.HashFunction.HetAtomTautomer)
'C=C[C@H]1C[N@]2CC[C@H]1CC2[C@H]([O])[C]1[CH][CH][N][C]2[CH][CH][C](OC)[CH][C]21_1_0'
>>>
>>> Chem.MolToSmiles(rdMolStandardize.Uncharger().uncharge(mol1))
'C=C[C@H]1C[N@@]2CC[C@H]1CC2[C@H](O)c1ccnc2ccc(OC)cc12'
>>> Chem.MolToSmiles(rdMolStandardize.Uncharger().uncharge(mol2))
'C=C[C@H]1C[N@]2CC[C@H]1CC2[C@H](O)c1ccnc2ccc(OC)cc12'

Expected behavior

rdMolHash and Uncharger should give the same output for mol1 and mol2.

Screenshots

Configuration (please complete the following information):

• RDKit version: 2024.03.1
• OS: Centos 7
• Python version (if relevant): Python 3.10
• Are you using conda? No
• If you are not using conda: how did you install the RDKit? Source install