RDKitToolkitWrapper, using rdkit=2024.09.3, writes some charged atoms to MOL/SDF without explicit charge

[j-wags]

Describe the bug

One very interesting thing I found while debugging this report is that, using the most recent version of RDKit, our RDKitToolkitWrapper no longer writes MOL blocks with formal charges assigned to (some?) positive nitrogens. It's likely related to this change (which I'm working to understand, so not clear this is a bug at all, and if it is, it's not clear which package it's in).

A simple example of this new behavior is:

from openff.toolkit import Molecule
temp = Molecule.from_smiles('[N+1]([H])([H])([H])([H])')
temp.to_file('temp.sdf', file_format='sdf')

Using rdkit 2024.09.3 this yields:

     RDKit          2D

  5  4  0  0  0  0  0  0  0  0999 V2000
    0.0000    0.0000    0.0000 N   0  0  0  0  0  0  0  0  0  0  0  0
    1.5000    0.0000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
   -1.5000    0.0000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
    0.0000    1.5000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
   -0.0000   -1.5000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
  1  2  1  0
  1  3  1  0
  1  4  1  0
  1  5  1  0
M  CHG  1   1   1
M  END
$$$$

But using rdkit 2024.09.2 we get

     RDKit          2D

  5  4  0  0  0  0  0  0  0  0999 V2000
    0.0000    0.0000    0.0000 N   0  0  0  0  0  4  0  0  0  0  0  0
    1.5000    0.0000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
   -1.5000    0.0000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
    0.0000    1.5000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
   -0.0000   -1.5000    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0
  1  2  1  0
  1  3  1  0
  1  4  1  0
  1  5  1  0
M  CHG  1   1   1
M  END
$$$$

Note the difference in charge flag (0 vs 4) in the first atom line.

The new behavior IS self-consistent - that is, our RDKitToolkitWrapper with rdkit 2024.09.3 can read the SDF that it wrote and get the same mol graph as it had originally:

from openff.toolkit import Molecule
temp = Molecule.from_smiles('[N+1]([H])([H])([H])([H])')
temp.to_file('temp.sdf', file_format='sdf')
temp2 = Molecule.from_file('temp.sdf')
print(temp2.to_smiles())

correctly yields [H][N+]([H])([H])[H]