Metrics for the Evaluation of Docking Model Quality

Thank you everyone for your help!

I have some more questions:

After obtained the docking results (Protein-DNA), I can see which docking structure is the best among all predicted structure.

(1)Do you think I could use Fnat and i-RMSD of the highest scored structure to support that our docking model is a high-quality one? my reference is this paper

(2) Empirically, how low a HADDOCK score can be considered as good? In my case, my best scored protein-DNA docking structure received a score of -83.4 +/- 6.9. Is it good enough?

Thanks in advance!

Pei

After obtained the docking results (Protein-DNA), I can see which docking structure is the best among all predicted structure.

(1)Do you think I could use Fnat and i-RMSD of the highest scored structure to support that our docking model is a high-quality one? my reference is this paper

Only if you know the reference structure of the complex you are predicting. Otherwise you can not calculate those metrics.
The HADDOCK results pages do show such stats, but in that case the reference structure used is the best model generated.

(2) Empirically, how low a HADDOCK score can be considered as good? In my case, my best scored protein-DNA docking structure received a score of -83.4 +/- 6.9. Is it good enough?

There is no absolute values to define what is good or bad. In principle the electrostatic energy can go to minus infinity :slight_smile:

What is good will depend on the type of the complex and the specifics of the molecules.

And remember: our HADDOCK scores IS NOT a free energy of binding.

Thank you amjjbonvin!

I only have
(1)a Alpha-Fold predicted protein structure
(2)a putative DNA binding site.
So, just to confirm:

  • in my case, Fnat and i-RMSD could not tell whether the best-scored docking structure is high quality or not.
  • Both two metric could do so only when there is some experimentally derived reference structure.

Am I right?

Thanks!

So, just to confirm:

  • in my case, Fnat and i-RMSD could not tell whether the best-scored docking structure is high quality or not.
  • Both two metric could do so only when there is some experimentally derived reference structure.

Am I right?

Yes - those metric in the result page plots allow you to assess the convergence of your run and see if the best scoring cluster also includes the best scoring model

Thank you amjjbonvin!
after read some literature, it looked like HADDOCK 2.2 was supposed to generate protein–DNA complex model(s) with available experimental data, like NMR spectroscopy.
If I only have predicted protein and DNA 3D structure, it looked like HADDOCK may not be very helpful in my case.
Am I right?
Thanks!

You can try to run HADDOCK in ab-initio mode (e.g. center of mass restraints), but without any information I would not trust much the models.
Unless you know which DNA sequence is being recognised, which would allow you to define a few active residues.

Also I do suggest to switch to using the 2.4 version of the server.

You can try to run HADDOCK in ab-initio mode (e.g. center of mass restraints), but without any information I would not trust much the models.

Yes, I have tried ab-initio mode.

Unless you know which DNA sequence is being recognised, which would allow you to define a few active residues.

I think we know. We have inferred roughly the DNA-binding domain and then we obtained the corresponding motif from JASPAR.

Also I do suggest to switch to using the 2.4 version of the server.

Sure. I used 2.4 since the very beginning. Cheers.
Thank you!

Hello! Has the issue with protein-DNA docking been resolved?

I’m currently working on a protein and DNA docking project as well, and I’m not sure under what circumstances HADDOCK results indicate stronger binding affinity. I look forward to your response. Thank you very much!

Which issue are you referring to?