Hi,

Yes, validation is required to publish an article. If most amino acids are involved in reconnection, it can be said that docking protocol is valid.

Otherwise you can do the following:

1- Dock again and do it correctly. Maybe the structures were not well prepared, or grid box is big and you gave the ligand a lot of freedom of action.

2- Use more accurate software for docking.

3- For better docking validation, you can do molecular dynamics and calculate free energy (MM-GBSA).

I agree with Farnoosh Kazemi that you need to validate before publishing. To dock the ligand I would:

1. get the experimental PDB structure of the protein from the PDB.

2. if the experimental structure includes the ligand, than you can just use that structure. If the ligand is not included than you should use docking software to dock the ligand to the experimental structure from PDB. There is different software out there, and what’s best for your case I can’t say. If the ligand is a small peptide/protein you may be able to use the RosseTTAFold server (or download the code, it’s free on github) to predict the protein-protein interface without docking. RoseTTAFold is a neural network for protein structure prediction similar to AlphaFold and does a decent job at predicting multimeric interfaces without docking (not good for antibody-protein complexes though).

3. to refine the model you can use the ISOLDE plugin for ChimeraX, which performs an interactive EM/MD to fit the coordinates of the model into the cryo-EM or x-ray maps. It’s okay if not every single atoms matches the structure exactly, but you definitely want to minimize RMSD between model and known structure, which ISOLDE can help with. Performing a short MD simulation can help with that as well as reduce clashes. If ligand isn’t included in force field, it needs to be parameterized, CHARMM-GUI should be able to handle ligand parameterization.

4. make sure to include any experimental data that is consistent with your model if that’s available when you write up your work.

Dear Farnoosh Kazemi and Ashley L. Bennett mam

Thanks for your help, could you please help me a little bit

My protein is hypothetical and it has no experimental structure in RCSB PDB. But I after Blastp I have selected the most close ortholog and also using ProBis I have selected the closest ortholog from PDB for that hypothetical protein. In that native PDB protein Coenzyme A (Ubiquinone) was the ligand.

Then I extracted the ligand from the native protein of PDB and redocked in my local machine using AutoDock Vina and counted the least binding energy and corresponding pose for that. One thing---***I used blind docking in that case.----Case1

Then I used the same ligand (CoenzymeA) and docked it with my hypothetical protein, and took the least binding energy pose into consideration. -----Case2

Then I generated 2D plot to understand ligand interactions, but it was showing different residues involved and just 2 or 3 residues were common but not at the same position, so 100% match was not there.

Then I superimposed both the complex ---the Native PDB-ligand and my experimental protein-ligand ----in PyMol setting the cycle undefined.

After 4 cycle it was showing the RMSD 2.3

Again I superimposed in PyMol but this time I defined cycle number at 0, then RMSD was 10.32...

Again I superimosed in an online server "SuperPose Version 1.0" there local RMSD was 1.53 and global RMSD was 11.66

Now what can I do, my protein is phosphopantetheneiyl transferase of Francisella tularensis but I took phosphopantetheneiyl transferase of Bacillus sp. is that making trouble??

What can I do now???

Please kindly help, thanks a lot

Sincerely--

Sunzid

Hello, you’re welcome.

Since your protein structure does not exist in PDB, I think you can use the closest ortholog. But you have to compare the involved amino acids in binding of the co-crystallized ligand and re-docked ligand, rather than comparison between two re-docked structures, and it is better to dock at the desired binding site instead of blind docking, but the size of the grid box should be appropriate and allow the ligand to move freely.

I agree with Farnoosh Kazemi that you should compare you docked ligand binding site to the experimental structure and that you should do the docking in the binding site. I would also add that a RMSD of 2.3 Å is actually not too bad. The width of a carbon atom is about 1.4 Å for a point of comparison. I would also expect that, because you’re using a different species for your docking, that there may be differences in the binding site. If there are any published mutagenesis experiments that show a specific residues effect on your ligand binding, you can use that to validate your model. You look for the residues shown as binding residues to interact with the ligand in your model. Also, be careful with RMSDs. There are backbone RMSD and all atom RMSD. Because you‘re looking at specific interactions between residues and ligand, you probably want to use an all atom RMSD, at least for the binding site.

Thank you, Farnoosh Kazemi and Ashley L. Bennett for the answers. I also have some questions and would like to get some clarification.

1) Should we use all C-alpha carbon atoms of the proteins or just the C-alpha carbon atoms in the binding site in order to get the all atom RMSD?

2) Is there any value of including other atoms like N and O atoms for the RMSD?

3) If the ligand is a peptide, should RMSD be calculated separately for the proteins and peptides? Is it still all atom RMSD in this case?

4) If there are any published mutagenesis experiments that show a specific residues effect on our ligand binding, how would you validate your model? Would you look at the 2D and 3D protein-ligand interactions and judge the quality of the docking based on the replication of the reported interactions in the docked complex visualised? Does in silico alanine scanning mutagenesis helps in any way for the model validation?

I look forward to hearing from you. Thank you.

Sicnerely yours,

Thai Leong

Thai Leong Lai my suggestions to your questions:

1) I would look at RMAS of overall protein being sure to get good RMSD in binding site. Sometimes people will report the full protein RMSD along with the RMSD for the binding site alone, so it may be worthwhile to actually calculated both.

2) Using backbone atoms for RMSD is usually a good approach, that’s what I typically do. Using C-alpha is also common so I think would be okay too. No need to include side chain atoms, as they are highly dynamic.

3) I would calculate the protein RMSD and also the RMSD of the peptide separately. This will ensure both the protein and peptide models are reasonable structures.

4) if there are specific residues known to interact with the ligand/peptide you would want to look for those interactions in your model. Good models reproduce experimental observations. I have had great success with in silico Alan in scanning mutagenesis so I do think it can be useful, especially if there is experimental data to compare to.

Hope this helps!

Thank you for the suggestions, Ashley L. Bennett