Re: [Rdkit-discuss] sampling of ring conformation for docking

[He, A. <he...@bu...>]

Hi Pavel,

Do you work with small rings (5, 6, 7) or large cyclic structures (like cyclic peptides)?
To distinguish different conformations of small rings, I feel that the torsional angles or apex heights – such geometric values that are alignment-free and depend on the internal coordinates of the molecules - might be more useful than RMSD. You can run conformation generation then put conformers into categories, if you don’t have too many rings and the rings aren’t that big. To get started with a simple example in RDKit, I previously found this tutorial very helpful:
https://sunhwan.github.io/blog/2021/02/24/RDKit-ETKDG-Piperazine.html

Docking in AutoDock Vina (https://autodock-vina.readthedocs.io/en/latest/) or AutoDock-GPU (https://github.com/ccsb-scripps/AutoDock-GPU) supports sampling of ring conformations on the run. By default, attempts will be made during docking to sample alternate conformers of 7-membered and larger rings. Optionally, you could also turn on the sampling for 6-membered rings and smaller ones. Take a peek at this recent paper to learn about the method:
https://www.cambridge.org/core/journals/qrb-discovery/article/performance-evaluation-of-flexible-macrocycle-docking-in-autodock/D8417BC284AEE198EC6AF25C7E677249

The Meeko project (https://github.com/forlilab/Meeko?tab=readme-ov-file#python-tutorial) provides a seamless workflow in Python to export your RDKit molecules into AutoDock-ready formats (and the docking outcomes can be retrieved back to RDKit, too!). The multiple docking outcomes with AutoDock Vina can give you at least some idea of what conformations might fit. You could refine the poses with more advanced methods.

Hope this helps!


Best regards,
Amy H.

From: Pavel Polishchuk <pav...@uk...>
Date: Monday, May 13, 2024 at 4:43 AM
To: rdk...@li... <rdk...@li...>
Subject: [Rdkit-discuss] sampling of ring conformation for docking
Hello,

   I use RDKit to embed initial conformations for docking. The issue is
with saturated rings. I can use a single random conformer but its
geometry may be unsuitable and the whole molecule will fail to dock. I
can use several starting conformers for docking and to avoid docking of
very similar conformers I can select a few diverse conformers based on
RMSD between rings only. However, the issue occurs if a molecule has
several such saturated rings. The current workaround is to compute RMSD
between corresponding rings individually, then average RMSD values and
select a diverse set of conformers. It may work to some extend.
   However I'm curious whether a better solution possible? Can we sample
rings individually and embed a molecule using pre-generated conformers
of some parts (rings)? I know about the restricted conformer enumeration
function, but it will work if we supply only a single connected part as
fixed. It should not work if we have two disconnected parts (rings) with
3D coordinates, because we do not know their relative position to
generate 3D coordinates for the rest of atoms in a molecule.
   Maybe someone will have some ideas/suggestions?

Kind regards,
Pavel


_______________________________________________
Rdkit-discuss mailing list
Rdk...@li...
https://urldefense.com/v3/__https://lists.sourceforge.net/lists/listinfo/rdkit-discuss__;!!KGKeukY!0p9-LhqopxbW2-tJTOxCwEVRUKO6jN5s_2WifPuV2PCrDjoa_nTmgY9NPdqsyDi2aHTJ3LA1_Kh37wI0Vhn8IlJ5PAEKr5vut811YA$<https://urldefense.com/v3/__https:/lists.sourceforge.net/lists/listinfo/rdkit-discuss__;!!KGKeukY!0p9-LhqopxbW2-tJTOxCwEVRUKO6jN5s_2WifPuV2PCrDjoa_nTmgY9NPdqsyDi2aHTJ3LA1_Kh37wI0Vhn8IlJ5PAEKr5vut811YA$>