Modelling help needed! Help us understand our matched pair data

[KangpingL]

Hi everyone,

In this project, we have a bunch of experimental data on the binding of small molecules to WDR91. Many of the molecules are in nice matched pairs, in which the molecules differ by a single, small change and yet have different binding energies (according to the SPR data we have obtained).

These matched pairs are shown in this deck. An example is below. We're reaching out for help in understanding these data.

We need computational modeling gurus to use the structure of the protein + a bound small molecule (shown in this paper, the starting point for the project) and try to figure out which interactions between the protein and the molecules we've measured are governing the binding.

i.e. What's going on?

If we can do this, we can design better binders towards a chemical probe for this protein.

Some preliminary work has been done by a Master's student at UCL, @Mohamadshahir15, and can be found if you click on "Issues", above.

This request has been shared to X and Linkedin

LInks to pdb files: WDR91-hit co-crystal structure https://www.rcsb.org/structure/8SHJ WDR91 covalent binder co-crystal structure https://www.rcsb.org/structure/8T55 apoWDR91 crystal structure https://www.rcsb.org/structure/6VYC

Hit:

Covalent binder:

The library of KDs of all screened compounds can be found in Summary of molecule

The SMILES of matched pairs with KDs can be found in Matched pairs with KD.xlsx This summary also includes the lastest Jan 2024 molecules.

The rest of the molecules in our screened compound lib can be found in #1.

[holeung]

Can someone please provide a file with the matched pairs and any other relevant molecules in SMILES or as an .sdf? Thank you.

[blackrabbit17]

I am keen to help on this too, please upload datasets and code / whatever you have so far

[suneelbvs]

Some thoughts on the impact of 'Phenyl group' : The crystal structure (PDB: 8SHJ) reveals that the phenyl moiety is situated within a hydrophobic pocket (refer the image), engaging in π-alkyl interactions with the Cys487 and Met535 residues. This positioning suggests that, phenyl group is well tolerated at this position.

:

[suneelbvs]

Follow up - Impact of phenyl group on SAR:

Aromatic substitution is well tolerated at this position, then aliphatic or heteroaryl rings.

[suneelbvs]

Some thoughts on the terminal aliphatic ring & its SAR:

[suneelbvs]

Follow up - thoughts on the terminal aliphatic ring & its SAR: .

Polar (acceptor groups) seems to be ideal substitution at this position, and even 5' and 6'-membered aliphatic systems tolerated well.

[KangpingL]

Hi everyone,

Thank you a lot for your help! The SMILES of Matched pairs with KDs can be found in Matched pairs with KD.xlsx This summary also includes the lastest Jan 2024 molecules.

The rest of the molecules in our screened compound lib can be found in #1.

[mattodd]

Hi @holeung @blackrabbit17 - hopefully that Excel is what you're after? @KangpingL - it might be good if we can provide a link (at the top of the page, in the original comment) to the pdb file(s) of the relevant structures here. The original and the covalent hit? Speeds up people being able to play with the data.

On LinkedIn there have been suggestions of sigma holes and SH-CF3 interactions. The latter is interesting from the point of view of our earlier attempts to target that buried SH for a double covalent probe.

[sherifelsabbagh]

I am open to helping. Should we put any result here or a new issue ?

[mattodd]

@sherifelsabbagh - great. It depends a little. No problem starting a new issue to explain what you're doing. Check to see if someone has done something related in another issue. Default: start a new Issue. You can connect relevant issues with the # symbol, e.g. if someone else is looking at the same matched pair. Ultimately issues are meant to be on defined questions or pieces of work and will ultimately be closed and folded into the Story So Far on the wiki.

But go ahead!

[sherifelsabbagh]

Last comment for today.

I tried to model this compound (with CH3 instead of OH) to understand why it showed no activity compared to the HIT. I docked the compound with same procedure and it shows a docking score of -7.1 (the HIT score was -7.4)

As shown here, no difference in conformation between both compounds. Compound with no activity had similar interactions to that of the hit. so where is the difference ? No idea.

I want to wait for anyone else to model this interaction. May be my approach is not good.

[suneelbvs]

Last comment for today.

I tried to model this compound (with CH3 instead of OH) to understand why it showed no activity compared to the HIT. I docked the compound with same procedure and it shows a docking score of -7.1 (the HIT score was -7.4)

As shown here, no difference in conformation between both compounds. Compound with no activity had similar interactions to that of the hit. so where is the difference ? No idea.

I want to wait for anyone else to model this interaction. May be my approach is not good.

@sherifelsabbagh: Binding modes seems fine to me. Have we explored other modifications at this position (like NH2, Ethyl, COOH?, NHMe, Ethyl, cyclopropyl), these modifications might provide us with additional insights about this position, whether hydrophobic or polar group tolerated at this position?.

[suneelbvs]

Last comment for today. I tried to model this compound (with CH3 instead of OH) to understand why it showed no activity compared to the HIT. I docked the compound with same procedure and it shows a docking score of -7.1 (the HIT score was -7.4) As shown here, no difference in conformation between both compounds. Compound with no activity had similar interactions to that of the hit. so where is the difference ? No idea. I want to wait for anyone else to model this interaction. May be my approach is not good.

@sherifelsabbagh: Binding modes seems fine to me. Have we explored other modifications at this position (like NH2, Ethyl, COOH?, NHMe, Ethyl, cyclopropyl), these modifications might provide us with additional insights about this position, whether hydrophobic or polar group tolerated at this position?.

@sherifelsabbagh: Is it possible that the oxygen group of the hydroxyl moiety forms hydrogen bonds with the Lys536 side chain? If so, it would explain why the hydroxyl moiety is active. If this interaction were lost (in the case of the methyl analog), it would be inactive? Is there any evidence of water-mediated interactions at this position in existing crystal structures? We can learn a lot about SAR by exploring in this direction.

[sherifelsabbagh]

Last comment for today. I tried to model this compound (with CH3 instead of OH) to understand why it showed no activity compared to the HIT. I docked the compound with same procedure and it shows a docking score of -7.1 (the HIT score was -7.4) As shown here, no difference in conformation between both compounds. Compound with no activity had similar interactions to that of the hit. so where is the difference ? No idea. I want to wait for anyone else to model this interaction. May be my approach is not good.

@sherifelsabbagh: Binding modes seems fine to me. Have we explored other modifications at this position (like NH2, Ethyl, COOH?, NHMe, Ethyl, cyclopropyl), these modifications might provide us with additional insights about this position, whether hydrophobic or polar group tolerated at this position?.

yes binding modes are fine but can not explain the difference in affinity ( or activity)..

I have no idea about other replacement... only this compound with CH3 instead of OH..

[sherifelsabbagh]

@suneelbvs the distance between the OH and the LYS536 is 6.9 A, so H-bond is unlikely to happen.

The crystal structure showed no water molecules near the ligand-OH.

[suneelbvs]

Last comment for today. I tried to model this compound (with CH3 instead of OH) to understand why it showed no activity compared to the HIT. I docked the compound with same procedure and it shows a docking score of -7.1 (the HIT score was -7.4) As shown here, no difference in conformation between both compounds. Compound with no activity had similar interactions to that of the hit. so where is the difference ? No idea. I want to wait for anyone else to model this interaction. May be my approach is not good.

@sherifelsabbagh: Binding modes seems fine to me. Have we explored other modifications at this position (like NH2, Ethyl, COOH?, NHMe, Ethyl, cyclopropyl), these modifications might provide us with additional insights about this position, whether hydrophobic or polar group tolerated at this position?.

yes binding modes are fine but can not explain the difference in affinity ( or activity)..

I have no idea about other replacement... only this compound with CH3 instead of OH.. Acceptor seems to be important for potency. Most of the top actives has acceptor at this position. Refer the SAR table from the publication: https://www.biorxiv.org/content/10.1101/2023.08.21.552681v1.full.pdf

[drc007]

Have you confirmed that the bought structure has the correct structure and absolute stereochemistry?

Also docking scores are not binding affinities. https://www.hindawi.com/journals/ijmc/2018/3829307/.

"It was shown that docking programs have a reasonable power to predict correct binding pose of the ligands. However, their scoring powers were not same for different protein families and also there is a weak correlation between docked scores and binding affinities of the ligands"

[suneelbvs]

@suneelbvs the distance between the OH and the LYS536 is 6.9 A, so H-bond is unlikely to happen.

The crystal structure showed no water molecules near the ligand-OH.

@sherifelsabbagh: Yes, I know. Lys536 is bit far way?, Any water-mediated interactions with Lys?. A superimposition of the three crystal structures shows that the loop conformation is flexible (refer the image). Molecular Dynamics simulations might help us to understand the active site flexibility and water-mediated interactions? at this position and also help us to understand this activity cliff better.

[KangpingL]

@suneelbvs @sherifelsabbagh Hi all, we screened some variation at the 3-hydroxypyrrolidine position previously
(0 KD for these 2)

and also in Jan

(This one is quite interesting - lower KD despite no OH/O)

which can be found in #1 #5

[suneelbvs]

Have you confirmed that the bought structure has the correct structure and absolute stereochemistry? Also docking scores are not binding affinities. https://www.hindawi.com/journals/ijmc/2018/3829307/. "It was shown that docking programs have a reasonable power to predict correct binding pose of the ligands. However, their scoring powers were not same for different protein families and also there is a weak correlation between docked scores and binding affinities of the ligands"

@drc007 : Agree with you. Thanks for sharing the article. @sherifelsabbagh Docking score may not be effective, especially when you're dealing with activity cliffs/Matching pairs.

[sherifelsabbagh]

Have you confirmed that the bought structure has the correct structure and absolute stereochemistry?

Also docking scores are not binding affinities. https://www.hindawi.com/journals/ijmc/2018/3829307/.

"It was shown that docking programs have a reasonable power to predict correct binding pose of the ligands. However, their scoring powers were not same for different protein families and also there is a weak correlation between docked scores and binding affinities of the ligands"

I completely agree with you that docking doesnt predict affinity correctly. So may be we focus on interactions involved rather than docking score.

[sherifelsabbagh]

@suneelbvs I can not see any water mediated interaction. I think MD simulations may be useful in this case because we cant rely on docking only.

[sherifelsabbagh]

@suneelbvs @sherifelsabbagh Hi all, we screened some variation at the 3-hydroxypyrrolidine position previously (0 KD for these 2)

and also in Jan (This one is quite interesting - lower KD despite no OH/O)

which can be found in #1 #5

this compound is very interesting since it has no OH group and the ring is bigger.

[holeung]

We could use some more quantitative analysis of binding free energy, like FEP (free energy perturbation). Will take a look in a few days when I have more time.

[sherifelsabbagh]

New Pair..

I docked the compound which showed very low Kd compared to the HIT. The compound has a peperidne ring instead of the 5-membered ring with OH group.

The compound has similar docking score, -7.4. The interactions are similar, however the bond distances are somehow shorter for the piperidine compound. This is obvious from the below figure where you can see a considerable difference in ligand conformation. This might indicate more favourable orientation leading to tighter interactions ??

[KangpingL]

We could use some more quantitative analysis of binding free energy, like FEP (free energy perturbation). Will take a look in a few days when I have more time.

Hi, I agree. Our collaborators have done some FEP calculations previously. They first validated the method against known actives and inactives and the predicted activity (scores) generally matched the experimental KD trend. They then ran it on 20 unpredicted molecules and their experimental results

It's quite interesting that the origin of KD=0.67 uM molecule is actually a dummy generated by FEP for transition between 2 "dissimilar" molecules

I will create a new issue for any update/ discussion of FEP to make this issue tidy and easier to read.

[KangpingL]

New Pair..

I docked the compound which showed very low Kd compared to the HIT. The compound has a peperidne ring instead of the 5-membered ring with OH group.

The compound has similar docking score, -7.4. The interactions are similar, however the bond distances are somehow shorter for the piperidine compound. This is obvious from the below figure where you can see a considerable difference in ligand conformation. This might indicate more favourable orientation leading to tighter interactions ??

Hi I have referred this to a new issue to make this issue easier to read. Could we continue the discussion on this new pair here #8 please? Thanks!

[holeung]

So FEP was successful with predictions whereas docking seems much less so. I have also seen several cases where docking works poorly but FEP works well. We probably want to use more rigorous physics-based modeling.

[sherifelsabbagh]

So FEP was successful with predictions whereas docking seems much less so. I have also seen several cases where docking works poorly but FEP works well. We probably want to use more rigorous physics-based modeling.

Is it okay to use something like mmgbsa restoring?

[holeung]

Give it a try! See how well it performs vs FEP on the same molecules.

On Sat, Mar 9, 2024 at 2:14 AM Sherif Elsabbagh @.***> wrote:

So FEP was successful with predictions whereas docking seems much less so. I have also seen several cases where docking works poorly but FEP works well. We probably want to use more rigorous physics-based modeling.

Is it okay to use something like mmgbsa restoring?

— Reply to this email directly, view it on GitHub https://github.com/StructuralGenomicsConsortium/CNP14-WDR91/issues/6#issuecomment-1986816022, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACTRXEUQ7OX2V6SGOZUTI43YXLOI3AVCNFSM6AAAAABEEB2TFKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOBWHAYTMMBSGI . You are receiving this because you were mentioned.Message ID: @.***>

[suneelbvs]

That's interesting. However, have you tried running MD simulations of this protein to understand binding site flexibility and/or importance of water role in this region. Since most of the MMPs (matched pairs) are single-atom modifications (such as H > Me > OH), strongly suggest the presence of waters or binding site flexibility at this region. In such cases, rigid docking will be little help, studying MD or WaterMap will actually serve us well at this stage, and this learning will also assist us to enrich FEP calculations. Let's hear from fellow compchem and medchem scientists.

[sherifelsabbagh]

That's interesting. However, have you tried running MD simulations of this protein to understand binding site flexibility and/or importance of water role in this region. Since most of the MMPs (matched pairs) are single-atom modifications (such as H > Me > OH), strongly suggest the presence of waters or binding site flexibility at this region. In such cases, rigid docking will be little help, studying MD or WaterMap will actually serve us well at this stage, and this learning will also assist us to enrich FEP calculations. Let's hear from fellow compchem and medchem scientists.

I didn't because I don't have resources unfortunately...let us see what others might say

[suneelbvs]

That's interesting. However, have you tried running MD simulations of this protein to understand binding site flexibility and/or importance of water role in this region. Since most of the MMPs (matched pairs) are single-atom modifications (such as H > Me > OH), strongly suggest the presence of waters or binding site flexibility at this region. In such cases, rigid docking will be little help, studying MD or WaterMap will actually serve us well at this stage, and this learning will also assist us to enrich FEP calculations. Let's hear from fellow compchem and medchem scientists.

I didn't because I don't have resources unfortunately...let us see what others might say

@sherifelsabbagh : You can try gromacs using google colab. Here is the repo: https://github.com/bioinfkaustin/gromacs-on-colab. Im happy to connect and discuss if you need any help or assistance. Pls. feel free to reach me out.

[sherifelsabbagh]

That's interesting. However, have you tried running MD simulations of this protein to understand binding site flexibility and/or importance of water role in this region. Since most of the MMPs (matched pairs) are single-atom modifications (such as H > Me > OH), strongly suggest the presence of waters or binding site flexibility at this region. In such cases, rigid docking will be little help, studying MD or WaterMap will actually serve us well at this stage, and this learning will also assist us to enrich FEP calculations. Let's hear from fellow compchem and medchem scientists.

I didn't because I don't have resources unfortunately...let us see what others might say

@sherifelsabbagh : You can try gromacs using google colab. Here is the repo: https://github.com/bioinfkaustin/gromacs-on-colab. Im happy to connect and discuss if you need any help or assistance. Pls. feel free to reach me out.

Thanks so much.. I know how to run grimaces on colab but I think running the simulation will take long time. may be some one can do it.

[mattodd]

Hi all. Since this issue is getting a little long, I'm just doing some curation, in case you see I've hidden some comments.

@holeung matched pairs now linked in original post. @blackrabbit17 let us know if you need more. @suneelbvs . Yes, little tolerance to variation of the middle aromatic ring thus far, and we wondered too about the value of the terminal OH in making hydrogen bond with residue or solvent.

LinkedIn suggestions:

Rodrigo Rodriguez and Gilles Ouvry suggested that cysteine 503 SH may be forming a bond with the para Cl or CF3 (a "sigma hole" type interaction, allowing a productive interaction between a halogen and an electron-rich residue). See this paper by John Spencer, who also commented that one must also take into account desolvation penalties for any apparently constructive interaction we might identify. John suggested that Cl can form halogen bonds and we could interrogate this through looking at other Cl isomers. The idea that there may be an interaction with a buried cysteine is interesting - @KangpingL was targeting that residue with a covalent warhead (e.g. an acrylate), but the chemistry is paused as it was quite challenging.

[holeung]

The binding site looks challenging to build a tight binder! The binding pocket is small, just enough to engulf the chlorophenyl. The chlorophenyl already looks perfectly matched to the pocket. CF3 is already too big. You might be able to get better binding by switching to bromophenyl. The rest of the ligand lies flat on the surface. This reminds me of the K-Ras binding site.

I think the only way to get much better binding is further pursuit of a covalent binder, similar to the case with K-Ras.

Have these compounds been tested for biological activity?

Attached are sdf files of the Matched Pairs dataset with some chemical properties calculated.

Matched_pairs.sdf.zip

[KangpingL]

The binding site looks challenging to build a tight binder! The binding pocket is small, just enough to engulf the chlorophenyl. The chlorophenyl already looks perfectly matched to the pocket. CF3 is already too big. You might be able to get better binding by switching to bromophenyl. The rest of the inhibitor likes flat on the surface. This reminds me of the K-Ras binding site.

I think the only way to get much better binding is further pursuit of a covalent binder, similar to the case with K-Ras.

Have these compounds been tested for biological activity?

Attached are sdf files of the Matched Pairs dataset with some chemical properties calculated.

Matched_pairs.sdf.zip

No, we haven't tested biological activity yet.