ETKDG and distance constraints | RDKit blog

[utterances-bot]

ETKDG and distance constraints | RDKit blog

Biasing the conformational search

https://greglandrum.github.io/rdkit-blog/conformers/exploration/2021/02/22/etkdg-and-distance-constraints.html

[sunhwan]

Is there a restraint for torsion? When I used ETKDG for a ring (piperazine), its conformation was distorted. I was looking for a restraint but couldn't find one. Any advice what to do in such case?

[greglandrum]

@sunhwan : no, the constraints can only be applied as distances.

To get better ring conformers, I would suggest trying out the srETKDGv3 parameters (described here: https://pubs.acs.org/doi/abs/10.1021/acs.jcim.0c00025).
You do that by picking the parameters like this:

ps = rdDistGeom.srETKDGv3() 

[sunhwan]

@greglandrum Thanks for the suggestion! srETKDGv3 does improve the results (but not quite perfect though). I collected a statistics: https://sunhwan.github.io/blog/2021/02/24/RDKit-ETKDG-Piperazine.html

In summary, ETKDGv1 generates ~30% chair whereas ETKDGv3 generates ~45% chair conformers.

[greglandrum]

@sunhwan: you really should AddHs() before generating conformers; that should dramatically improve the quality of the comforters you get.

[sunhwan]

@greglandrum Thanks! My bad I forgot to add AddHs. It does improve the result. I have updated my post.

ETKDGv1 generates ~40% chair whereas ETKDGv3 generates ~72% chair conformers.

[chem555]

@greglandrum Hi, are the conformers generated from ETKDGv3 already geometrically optimised (MMFF) please?

[greglandrum]

@chem555: They are not. That's a step you need to do yourself if you think it's important.
The conformers which come from ETKDG tend to be pretty good, so you may not need that at all.

[chem555]

@greglandrum : Thanks for the clarification!

[ssrb19]

@greglandrum I am following your example, but when I switch to my molecule, I get the following error: Violation occurred on line 55 in file Code/DistGeom/DistGeomUtils.cpp Failed Expression: ub >= lb I am not sure what this means.

[greglandrum]

@ssrb19 in order for me to be able to help here you need to share the molecule and constraints that you used