Why are hydrogens needed for generating conformer (EmbedMolecule) ? And what are the risk to generate a conformation without it ?

[vincrichard]

Hello rdkit community,
My question arises from an issue I am facing when generating a constrained conformer on a smiles with stereochemistry. I can't share the structure for confidential reasons but it can be assumed to be correct.
I have a core which is used for constraining the embedding. The molecule itself has no problem with conformation generation.

I notice that the constrained generation fails when I generate while adding hydrogen, but works if I just pass the molecule as it is.
When debugging, I found that FINAL_CHIRAL_BOUNDS was the main cause of error in the EmbedMolecule method, and that passing enforceChirality to False make it possible to generate conformers.
However, I am not sure of the implications of both cases.

For now I would be inclined to generate the conformer without H, add H after conformer generation (with coordinates), and optimize the conformation with UFFGetMoleculeForceField (with a constrain on the core).
Again, I am wondering how much error it can involve and what are the risks to proceed this way ?

Any advice ? Or pitfalls I should be aware of?

[vincrichard]

I also notice that on an older version of rdkit 2022.09.1 (the previous reported information is based on the last release 2023_09_05). The embedding fails when Hs are not added to the molecules. I wonder what type of modification has been made to EmbedMolecule between those version ?

[vincrichard]

I traceback the cause to this issue / fix: #5883.