Issue with stereochemistry - RDKit seems to generate wrong sdf representations (correct in SMILES)

I have created a workflow for aligining molecules in 3D. A user (chemist) noticed an issue that in some cases the stereochemistry in the alignment is different from the input structure. The worflow uses the "Generate Coordinate", "Add Conformers" and "Optimize Geometry" Nodes.

I investigated the issue and made following observations:

When copying the cell containing the RDKit molecule, SMILES is copied even when using the SDF-String renderer and an SDF-string is displayed. When pasting that smiles into Chem3D the stereochemistry is correct, eg. the same as the input structure.

In contrast when using the Interactve BioSolveIT Viewer the stereochemistry is wrong, eg. different from the input for the exact same row (structure)! If I then use the SDF Writer and load the sd-file into Chem3D the stereochemistry is also wrong!

My conclusion is that the RDKIt nodes mentioned above to correctly keep stereochemistry the same as the input but there seems to be a bug when converting the internal representation to an sd-.file with 3D coordinates!

Please find attached the input structure (Androstenone) and 1 conformation (Row1693) as mol file with wrong stereochemistry and as SMILES with correct stereochemistry (when rendered in Chem3D). They are in the attached zip file as .mol extension is not supported for upload.

ADDITION:

Just to be sure I also checked the SMILES in Avogadro2 and the stereochemistry is wrong too and different for the same SMILES in Chem3D!  So I'm unsure what the actual issue is. See the 2 attached png images.

And then on

https://www.molecular-networks.com/online_demos/corina_demo

if I paste the SMILES for Row1693 it converts it to a molfile with the correct stereochmeistry, eg. same as inpout and same as Chem3D (this mol file also has correct stereochemistry when viewed with Avogadro).

The important part is that after 3D Coords and conformation generation the stereochemistry should not be changed. And I do think there is an issue in RDKit with the conversion from SMILES to SDF.

 

 

 

 

I can confirm that this is a bug in the 3D coordinate generation. Sorry for the inconvenience.

This sometimes happens in highly constrained molecules like this one.

I'll file the bug and take a look at trying to fix it, but I can't make any promises.

-greg