For the peptide docking all my values have been negative in the range of -77 to -111. I had just assumed that the haddock score was an indicator of binding affinity. I have done a lot of peptide docking and have been comparing the haddock scores with each other to determine which could have a potentially better binding affinity. So instead of just doing this I should deduct 10% of the restraint energy from each score then do a comparison? For all my peptide dockings I have applied the same restraints. Would this still require deducting 10% of the restraint violation energy?
And thank you for your help on my questions it is much appreciated.
For the peptide docking all my values have been negative in the range of -77 to -111. I had just assumed that the haddock score was an indicator of binding affinity.
When comparing similar systems based on similar restraints, it might correlate.
I have done a lot of peptide docking and have been comparing the haddock scores with each other to determine which could have a potentially better binding affinity. So instead of just doing this I should deduct 10% of the restraint energy from each score then do a comparison? For all my peptide dockings I have applied the same restraints. Would this still require deducting 10% of the restraint violation energy?
If the restraints are similar, then it should be fine. Although removing it would give you a cleared energy picture of the interaction
Is there any significance of removing just 10% of the restraint violation energy instead of 20% for example? Also, are the RVE’s expected to be close to 0? I have tried to read on the significance of the RVE in docking but I don’t know if I fully understand yet. Mine are usually over 100 and I don’t know if that impacts the accuracy of my docking?
Also in terms of the summary it gives how many structures were clustered (in to how many clusters) and what percent of water refined models it generated but I am not fully sure on what these mean and what would be good to use for comparing my different peptides as potential inhibitors.
Thank you
The 10% is the default weight of the restraint energy in the HADDOCK scoring function.
So unless you have modified it, this is what you should use to subtract the restraint energy.
Is there any significance of removing just 10% of the restraint violation energy instead of 20% for example? Also, are the RVE’s expected to be close to 0? I have tried to read on the significance of the RVE in docking but I don’t know if I fully understand yet. Mine are usually over 100 and I don’t know if that impacts the accuracy of my docking?
The restraint energy will depend on the quantity and quality of the restraints you defined.
Also in terms of the summary it gives how many structures were clustered (in to how many clusters) and what percent of water refined models it generated but I am not fully sure on what these mean and what would be good to use for comparing my different peptides as potential inhibitors.
Cluster size is not used as a scoring/selection criteria.
I have made no modifications to the default settings for the restraint energy. And all conditions such as active residues and restraint modifications are the same. With this in mind I can make the 10% deduction, but I don’t know how necessary it is in my case.
Ok so Haddock score solely should be used when evaluating potential inhibitor affinity?
Also, how different can a different of one for Haddock score be perceived? For example if for one peptide it is -110 and another it is -111. Is a difference of one still considered significantly different. Or would it have to be greater. I am using this as a way to predict which peptides I am developing would be better inhibitors. So want to choose those with the best score to synthesize but want to know what difference is considered significant.
thank you
Most likely not - look at the standard deviation of the scores when comparing those.
E.g. -111±5 is not different statistically different than -110±4 …
Ok so can I compare haddock scores if the restraints are the same as well as using the same protein. The only thing being changes is the peptide I am docking to it? Next, I will be using the same peptides to dock again but I will be changing some of the active residues on the protein it is being docked to.
What does it mean to have a high restraint violation energy, is it better for it to be smaller?
I know you have mentioned before that there is no correlation between docking scores and binding affinity. But in my case I am docking different peptides to the same protein, and I will keep active residues the same. By comparing my scores can I assume those with a better score demonstrate a potentially better binding affinity?
I have read some posts on the forum and want to know whether refinement or prodigy is necessary for my case? My goal of using haddock is to predict which peptides I develop will have a better binding affinity to the protein of interest.
Lastly, my peptide is bicyclic and has a disulfide bond. When I check the PDB structure of the complex, I notice that the peptide isn’t cyclized, with the head to tail cyclization. So although it looks cyclic, it isn’t cyclic. I am not sure what causes this as the pdb file I submit is a closed cyclic structure with the disulfide bond present.
Thank you for your continued help
What does it mean to have a high restraint violation energy, is it better for it to be smaller?
If your model satisfies all the restraints (the info given to HADDOCK), the restraint energy should be 0.
So the smaller the better.
I know you have mentioned before that there is no correlation between docking scores and binding affinity. But in my case I am docking different peptides to the same protein, and I will keep active residues the same. By comparing my scores can I assume those with a better score demonstrate a potentially better binding affinity?
As a first approximation yes, but again score is not equal to binding affinity.
I have read some posts on the forum and want to know whether refinement or prodigy is necessary for my case? My goal of using haddock is to predict which peptides I develop will have a better binding affinity to the protein of interest.
prodigy is just another predictor of binding affinity. It was trained for protein-protein binding affinity prediction and not peptide.
Refinement (e.g. using the haddock server) could be done to remove the restraints contribution.
Lastly, my peptide is bicyclic and has a disulfide bond. When I check the PDB structure of the complex, I notice that the peptide isn’t cyclized, with the head to tail cyclization. So although it looks cyclic, it isn’t cyclic. I am not sure what causes this as the pdb file I submit is a closed cyclic structure with the disulfide bond present.
It the conformation you are giving to HADDOCK cyclised? If that it the case then you need to tell HADDOCK it is a cyclic peptide.
This is an option in the first submission tab (only appears if you have elevated access to HADDOCK which you can request in your registration page).
My restraint energies are mostly always over 100, even for my best peptide with a HADDOCK score of -132. Could having active residues on the protein which my peptide doesn’t interact with be a factor? I found based on the end structures that my peptide usually binds in a certain location and so some of the active residues I specify don’t make contact in the docking. Or should this not be a factor? Would a high restraint energy mean the haddock score or docking be inacurate?
I want to employ this as part of my Master’s project. Would you say this would be a good tool when determining inhibitor potential? And would there be a better alternative to call it than an evaluation of binding affinity?
Yes the conformation I give to haddock is cyclized and I have told it that it’s a cyclic peptide in the first tab. It’s just when I import it into chimeraX that I see it isn’t a closed cyclic conformation.
Non-zero restraint energy indeed indicate some of your residues are not contacting the peptide.
Which is not per se a problem, but the reason you should subtract it from your scoring when comparing your peptides
Yes the conformation I give to haddock is cyclized and I have told it that it’s a cyclic peptide in the first tab. It’s just when I import it into chimeraX that I see it isn’t a closed cyclic conformation.
That could be simply a viewing problem.
Another software like PyMol might display the bond.
I changed some of the active residues on the the protein, as based on one of the complexes produced I was able to determine which residues my peptides seem to be bonding to (evaluating H bonds in chimeraX). After making these changes the haddock score improved a little from -132 to -136, but the restraint energy went right down from 134 to 4.5. So I suspect it was the fact that I had imputed active residues on my protein where not all were interacting with my peptide. Does this seem a more reasonable number?
Clearly lower. But again, simply remove 10% of the restraint energy from the score to compare your peptides.
If they are that low won’t make much different though.
The peptides I have currently docked were altered with ChimeraX from a published PDB. But I want to make a comparison to the group of peptides I have currently docked, where there is not yet a PDB I can modify. I was going to make just the linear peptide but will not be able to cyclize it as intended, with any software. If on the first tab of the HADDOCK2.4 submission I put that it is a cyclic peptide. Will it then allow it to be cyclic for the purpose of analysis or does the PDB already need to be the cyclic peptide?
Thank you
You need the termini to be close in space for the server to recognise it as a cyclic peptide.
Check our paper on modelling protein-cyclic peptide complexes.
In there we use a local version of HADDOCK to generate cyclic conformations
Thank you, I will try to work on this. I am curious why did you not use SFTI-1 PDB to modify your peptides. Was there any advantage from using PyMol to generate the linear sequence? Instead of just modifying a PDB available?
Also, I changed the active residues on my target protein based on some of the docking information and visualizing the bonds my peptide makes to the protein. When I changed the active residues, I noticed my haddock score changed and my restrain energy lowered. But with regard to the haddock score, some peptides didn’t change much but some did a lot. So the rank of my peptides based on the haddock score changed. Why would this be? And would my change of active residues be considered more accurate?
Thank you
Thank you, I will try to work on this. I am curious why did you not use SFTI-1 PDB to modify your peptides. Was there any advantage from using PyMol to generate the linear sequence? Instead of just modifying a PDB available?
To be unbiased.
Also, I changed the active residues on my target protein based on some of the docking information and visualizing the bonds my peptide makes to the protein. When I changed the active residues, I noticed my haddock score changed and my restrain energy lowered. But with regard to the haddock score, some peptides didn’t change much but some did a lot. So the rank of my peptides based on the haddock score changed. Why would this be? And would my change of active residues be considered more accurate?
Again to compare different peptide subtract 10% of the restraint energy from the score.
I have been altering peptides from SFTI-1 PDB by altering certain residues (selecting rotamers) using chimeraX. What biases would there be from this?
I have been trying to follow the protocol from you paper (to see if modelling the peptides this way rather than altering the SFTI-1 PDB will make a difference. I have been made a linear sequence using PyMol (Antiparallel beta sheet). But I am not sure how you use Haddock2.4 to bring the termini closer? It wouldn’t be the same place as the docking submission?
I also repeated one of my runs with the exact same peptide, protein and restraints but the haddock score changed from -136.4 to -133.9. What would be the cause of this?
Thank you
I have been altering peptides from SFTI-1 PDB by altering certain residues (selecting rotamers) using chimeraX. What biases would there be from this?
We were doing benchmarking on set of diverse complexes.
You are working on a specific system and your approach should be perfectly fine.