Guidance on Using CReM Dock for Multi-Site Molecule Design

[28vipraparekh]

I am currently trying to incorporate CReM Dock into my master's dissertation research in Bioinformatics. I have a few technical questions regarding its implementation for my specific research needs:

1. I have 3 adjacent sites over my target, I am willing to design a molecule which can address three of them. And also I have a co-crystal molecule which is big enough to cover these three sites simultaneously.So, what should be my approach to use Crem-Dock : i. De-Novo drug design or ii. Fragment Expansion? 
If I go for fragment expansion, this approach will not suit my purpose as my co-crystallized molecule is already big and occupying all the sites. But at the same time I do not want to lose the existing inhibitor information.
If we go for de novo drug design then how can we include the information regarding existing inhibitors?
2. Is there a way to create a fragment library using these existing inhibitors or do we always use the precompiled fragment library by CREM?
3. In the tutorial, 3 fragment libraries are mentioned .How to decide which library to be selected ?
4. Can we keep interacting water molecules in the active site while performing Crem-Dock?

I would greatly appreciate any insights or guidance on these questions. Thank you for your time and for developing this valuable tool!

Best regards,
Vipra Parekh

[DrrDom]

1. There are several possible strategies.

a) you may use some starting fragments, which you may choose at your own (e.g. fragment known binders), or take the starting fragments from the SI of the preprint - https://doi.org/10.5281/zenodo.14577996 (de novo mode)

b) you can fragment a co-crystallized ligand on multiple fragments and use them as starting ones (de novo mode)

c) you can cut a particular fragment from a co-crystallized ligand and use it as a starting one, you can do this multiple times selecting each time a different fragment of a co-crystallized ligand and run in fragment expansion mode (do not change initial coordinates of the starting fragment in this case)

2. Sure, you can create an own CReM fragment database from any set of molecules, see instructions https://github.com/DrrDom/crem?tab=readme-ov-file#generation-of-a-fragment-database. The only issue which may occur, if the fragment database will be to small and no replacements can be found with a larger radius. However, this can be established only after an experiment, so you may try

3. There are two libraries with Csp3-enriched starting fragments. So, if you want to design more Csp3-rich structure, use one of them. "sa25" is larger than "sa2", which is a subset of the former set. If you do not bother about the fraction of sp3 carbons in generated structures you can use the third set as starting.

4. We did not try that. However, this will completely depend on a docking program. I expect, that if you will keep water molecules as a part of a protein vina and gnina should handle this correctly. You may try to do that. If some issues will happen we can discuss them later

A general advice is to use large radius if you want to generate really synthetically feasible structures with higher chances. I suggest radius 3 as a minimum.

I uploaded all three sets of starting fragment to zenodo.

[28vipraparekh]

Thank you for your guidance! I will explore all the suggested options and reach out if I encounter any difficulties.

[28vipraparekh]

Hello,

We decided to go ahead with the fragment-expansion for the purpose of our project . The script runs successfully; but, we have encountered an issue:

1. Our goal is to design a molecule that can cover three adjacent binding sites simultaneously. However, based on our results, the generated molecules only cover two sites at a time, rather than all three. Could you provide insights into why this might be happening and any possible solutions?

2. Additionally, we were advised to set the radius threshold to 3 in the above response. Could you clarify how exactly this threshold should be defined and implemented within the script?

[DrrDom]

1. It is not possible to guide growing in a specific direction. Growing is guided by docking. There are several reasons why you do not observe molecules filling all pockets:

• the chosen molecular weight or other physicochemical parameters are too restrictive; generated ligands are small and cannot physically fill all three pockets. In this case increase these parameters accordingly.
• there is no possibility to grow ligands in a direction of the third pocket. It may happen, but it is unlikely. If so, you may choose Pareto selection strategy to get more diverse solutions and additionally you may increase the number of replacements (max_replacements argument) to get a greater number of compounds on each iteration.

2. radius is a command line argument, which controls which bonds can be created during fragment growing. A larger radius makes growing more conservative and the output molecules should be more synthetically accessible, but the number of compounds will be smaller. You may check these papers: https://doi.org/10.1186/s13321-020-00431-w and https://dx.doi.org/10.1021/acs.jcim.0c00792 and the preprint of CReM-dock itself, there we investigated the effect of the radius. Radius 3 is quite safe to get a large fraction of feasible molecules.

[28vipraparekh]

Okay, Thank you very much for the suggestions and help!

[DrrDom]

I was not fully correct in my previous. There is a way how to guide generation. This is a hidden option, which I forgot to add to README. I'll update it.

You can supply input structures as 3D molecules which were pre-docked or taken from x-ray structure. It is important that their poses should be already pre-aligned to the target. For those 3D molecules in an SDF file you may provide a field named "protected_user_ids" with atom indices separated by comma (e.g. 1,2,5,6). Indices are started from 1. Those atoms will not be expanded anymore on future steps.

So, in your case you can (1) select some molecules, which already targeted two pockets, (2) protect atoms, which are far away from the third pocket, and (3) run a separate expansion round for these 3D molecules. We did not test this strategy with multiple input compounds, but it must work. There are no internal limits on that.

To further restrict growing, I recommend to additionally specify --rmsd threshold. In this case only those new structures, which are aligned well with a parent molecule, will pass to the next iteration. Thus, you will make protection of individual atoms more relevant, because otherwise a child molecule can flip in a pocket and growing will continue in an undesired direction.

[28vipraparekh]

Thank you! I'll give this option a try and see if I get any significant results.