Capri Criteria For Evaluating Docked Protein Complexes

Question

[Reyhaneh]

Hi;

Critical Assessment of PRedicted Interactions (CAPRI) has specific criteria for ranking the protein complex predictions. These criteria are :

1- 'fnat' "the number of nativeresidue–residue contacts in the predicted complex divided by the number of native contacts in the target"[1].
2- L-rms "is the backbone root-mean-square displacement of the ligands in the predicted versus the target structures"[1].
3- I-rms, "is the root-mean-square displacement of thebackbone of the interface residues only, in the predicted versus the target complexes"[1].

I am looking for a tool (or a script) which can take a .pdb file of a protein complex and analysis it based on CAPRI criteria and output the three main criteria. Do have any suggestion?

Thanks for you help.

reference: [1] "Assessment of Blind Predictions of Protein–Protein Interactions: Current Status of Docking Methods".Mendez et. al. PROTEINS: Structure, Function, and Genetics 52:51–67 (2003)

[Keith Callenberg]

Did you try contacting the CAPRI judges? They must have written a script for this already. Here is their contact info: http://www.ebi.ac.uk/msd-srv/capri/management.html I have emailed them before and they are fairly quick to respond.

[Reyhaneh]

Yes I have. They mention that the toll is not publicly available. although they helped me to process some of them but I need to develop myself for long term use.

[MG90]

Hello, I saw your question. And I kind of have the same question. did you figure out how to assess the protein-protein Docking output according to the CAPRI criteria? Would be really helpful to hear from you.

Answer 1

[João Rodrigues]

In our lab we use Profit (http://www.bioinf.org.uk/software/profit/) for the ligand and interface RMSD calculations. Depending on the complexity of your system this might be easy or not at all. The steps we take are:

1 - Define an interface region between both chains in the native structure. Basically, check for all atoms of chain A which atoms of chain B are within 10Å. This gives you a list of residues that are the interface for your complex. You can use Biopython (http://biopython.org/wiki/Main_Page) and its NeighborSearch module, for a free and simple solution for this calculation.

2 - For two complexes, with one being the native, a simple script to calculation a fraction of contacts between these lists gives you FNAT.

3 - For interface RMSD, use this list of contacts to define ZONEs in profit. For example, if you find your interface is composed of residues 1,2,3,4,5,55,90,99,100, for chain A, you can define this:

REFE native.pdb MOBI model.pdb ZONE A1-A5 ZONE A55-A55 ZONE A90-A90 ZONE A99-A100 FIT

This FIT command will first align the structures and then calculate the RMSD, both only over those particular residues. This is the iRMSD.

4 - For ligand RMSD, the same rationale works, but since the point is to calculate the deviation of the ligand only, you first have to fit on the receptor and then NOT FIT on the ligand but only calculate the RMSD. Again, in Profit, and assuming chain A is the receptor and chain B is the ligand:

REFE native.pdb MOBI model.pdb ZONE A FIT ZONE CLEAR RZONE B RMS

The second value to be outputted is your ligand RMSD.

Oh, and since it's only backbone, you can define that in profit with: ATOMS CA,C,N,O

---

While Pymol and VMD might give you approximate answers, Profit is much faster and more accurate. Also, it deals much better with exotic ligands and RNA/DNA. I found that pymol fails in this kind of complexes. Hope this helps!

[Reyhaneh]

Dear @Joao, thanks for this complete description. I have another question as well. I am docking two unbound-unbound structure such as (2cpl-A and 1e6j-P) and I know that my target should be a complex in PDB (e.g. 1ak4-A-D). in this case to for calculating FNAT for example do I need to superpose my unbound structures on the native? thanks;

[João Rodrigues]

For FNAT there is no need to superimpose the structures. You just need to have two complex structures, list each other's contacts, and compare the two lists. A couple of things to be aware of:

1. Calculating an FNAT between different structures doesn't make much sense either. They should be different conformations of the same structure.
2. Make sure the numbering is consistent between the structures. Otherwise the contacts might not overlap as they should and you have wrong results.

[João Rodrigues]

For FNAT there is no need to superimpose the structures. You just need to have two complex structures, list each other's contacts, and compare the two lists. A couple of things to be aware of:

1 - Calculating an FNAT between different structures doesn't make much sense either. They should be different conformations of the same structure.

2 - Make sure the numbering is consistent between the structures. Otherwise the contacts might not overlap as they should and you have wrong results.

[filippis]

For the iRMSD, is not necessary to define also the interface residues for chain B?

And if native and model do not have same chain labeling and res numbering how can someone use the ALIGN command? I mean that alignment should be used to map receptors and do the fitting and also for mapping ligands and calculating the ligand-RMSD.

Many thanks for the help Joao

[João Rodrigues]

The problems of numbering and correct chain naming is a huge one.. I recommend you do that manually for each case, automatically might be complicated (sequence alignment, etc etc).

The iRMSD takes into account both receptor and ligand(s) yes.

[ramachandranakash1994]

@ João Rodrigues - I am new to programming and do not understand the working of NeighborSearch. Would you be able to give me the code to find all of the interface residues (as defined by you) between 2 chains in heterodimeric proteins?

Answer 2

[Keith Callenberg]

Since the CAPRI website doesn't seem to have any scripts already available, I think this would not be too difficult to build yourself.

You could pretty easily write a script to use PyMOL (python) or VMD (TCL) to do this.

Here is a quick guide to scripting in PyMOL: http://www.pymolwiki.org/index.php/Simple_Scripting

And there are a bunch of VMD script examples that you could start from here: http://www.ks.uiuc.edu/Research/vmd/script_library/

[Reyhaneh]

Thanks for this.

Answer 3

[Houcemeddine Othman]

The MMTSB tool set could be handful. it contains a script for identification and comparison of native contacts. http://blue11.bch.msu.edu/mmtsb/Main_Page

otherwise you could use PROFIT to calculate L-RMSD and I-RMSD. for I-RMSD you could use the ZONE function to define your interface residues.