[BUG] Incorrect charges from Sire to RDKit conversion

[tom-potter-cresset]

Describe the bug
Converting some molecules from Sire to RDKit results in some strange formal charges and bonded structure. For example, the following molecule:

becomes (formal charges labelled):

This only happens if loading from a format which lacks bond-order information (eg. prmtop). Loading from an sdf, for example, works fine.

To Reproduce
Steps to reproduce the behavior:

mol1 = sr.load('input.inpcrd','input.prmtop')[0]
rdmol1 = sr.convert.to_rdkit(mol1)
for atom in rdmol1.GetAtoms():
    print(atom.GetFormalCharge())

mol2 = sr.load('input.sdf')
rdmol1 = sr.convert.to_rdkit(mol2)
for atom in rdmol2.GetAtoms():
    print(atom.GetFormalCharge())

I'm not sure what the ideal solution would be, as the information about the correct structure isn't available. Possibly there could be some way to flag unlikely groups like carbanions?

inputs.zip

[lohedges]

In the absence of stereochemistry, the converter uses an inference routine based on one from MDAnalysis. This works well in many cases, but clearly isn't ideal for everything. I'll take a look at your example tomorrow in case there is a bug somewhere. There have been a few subtle issues in the past.

[lohedges]

I also think the fallback relies on appropriate 3D coordinates, so might not work as intended for structures that aren't well minimised.

[tom-potter-cresset]

I just checked, and I get the same result when I minimise the molecule first

[lohedges]

Thanks for the update. I've tried using some of the RDKit stereochemistry inference routines, but those fail to assign the correct formal charges too. I'll take a look at some of the sanitization options to see if disabling any helps in this case.