Difference in InChIKey values, from SMILES, when comparing to RDKit

If I run the following with RDKit (2021.09.2):

from rdkit import Chem    

m1 = Chem.MolFromSmiles("N1C[C@@]2(CC[C@H](C)CC2)CC1")  
print(Chem.MolToInchiKey(m1))    

m2 = Chem.MolFromSmiles("N1C[C@]2(CC[C@H](C)CC2)CC1")  
print(Chem.MolToInchiKey(m2)) 

I obtain:

SGJOFPWLAFWQEP-AOOOYVTPSA-N
SGJOFPWLAFWQEP-MGCOHNPYSA-N

If I do something similar with OpenFF’s to_inchikey() method (openff-toolkit 0.10.1)

from openff.toolkit.topology import Molecule    

m1 = Molecule.from_smiles("N1C[C@@]2(CC[C@H](C)CC2)CC1")  
print(m1.to_inchikey())      
 
m2 = Molecule.from_smiles("N1C[C@]2(CC[C@H](C)CC2)CC1")   
print(m2.to_inchikey()) 

I obtain:

SGJOFPWLAFWQEP-UHFFFAOYSA-N
SGJOFPWLAFWQEP-UHFFFAOYSA-N

Any ideas why? I am missing something?

Many thanks,

Mark

[j-wags]

Per the discussion above, it looks like a new feature in RDKit CAN handle this.

I’ll start poking around in OpenEye as well. One hint is the yellow warning in the Omega docs that ring stereo may not be respected.

[j-wags]

I’ve started a discussion on the RDKit board here: https://github.com/rdkit/rdkit/discussions/4731