You are here

Problems with the convergence of docking by the graph

5 posts / 0 new
Last post
Problems with the convergence of docking by the graph
#1

I have been doing the docking with the program Rosetta
for an antigen-antibody complex
repeated runs and not get a graph
funnel-shaped
because the lowest energy structure
have a higher rmsd 5Ǻ
wanted to know if that's possible
really appreciate the attention you can give me.

PS
attached a file with the results in pdf format

In these cases it will be worth re-do the docking from some of the results, ie, do a re-docking of the structure obtained in order to obtain a better convergence?

AttachmentSize
docking Job.pdf326.43 KB
Post Situation: 
Sun, 2012-03-25 04:56
Kanorgon

What protocol are you using? You posted this in both Rosetta3 and Rosetta++.

Perturbational re-docking of a docking model will probably sample fairly widely. Local refinement docking (docking_local_refine) will vary a lot less. I would guess you'll get a tighter cluster of models this way but I wouldn't expect RMSD to fall.

What is the reference structure that your models are 5 Angstroms from? Are you trying to get Rosetta to demonstrate that its scorefunction is reasonable by reproducing a crystal structure? Is this blind docking of a new complex?

Sun, 2012-03-25 08:27
smlewis

good afternoon
Thanks for answering!
Actually I used the old protocol Rosetta + +.
I've been doing the docking to a new antigen-antibody complex and according to the reference of Gray J. et.al. 2003, "To Quantify the
Presence of the funnel, we examine the five lowest-scoring
decoys' for each target. If three or more of
Those decoys are within The 5A ° rmsd of the native
structure, we say That the target exhibits a score
funnel ".

Sun, 2012-03-25 13:16
Kanorgon

One of Steven's points was that rmsd is only a useful metric if there's an appropriate reference structure you can compute the rmsd to. In practice this means rmsd is only useful if you already have an experimentally determined structure of the complex you're trying to model.

One way to validate computational modeling is to take the unknown complex you want to model, and pick a few related examples with known experimental answers. You can then run the protocol the same way on both the unknown complex and the related, known complexes. You can then check how close the protocol gets to the known experimental structure on the knowns, and assume that the protocol did about as well on the unknown complex. In essence, you're using the known complexes as positive controls. It's for analyzing these positive controls that the rmsd metric is reported.

Rmsd is of limited, if any, usefulness when you're doing prediction of new complexes. In that case you'd actually expect there to be some rmsd differences between any existing structure and the newly modeled complex - if you don't expect there to be any atom movement, why do modeling? You should expect some movement, although if there's too much, something might have gone wrong. (I'm not familiar with antibody modeling, so I don't know if 5 Ang or even 20 Ang is a reasonable difference or not from one complex to another complex.)

Are you doing modeling of a complex with a known structure (e.g. a positive control), or are these results from an unknown complex? In both cases, what are you using for the "correct" reference complex, and how are you specifying it?

Sun, 2012-03-25 14:53
rmoretti

Thanks for your reply!!!
If, indeed I modeling antigen-antibody unknown complex
We conducted a previous validation of the method of docking with Rosetta by using a estructure of antigen-antibody cristalline
at that time got the graphic with low rms values regarding the reference structure
Now I understand, Thank for you clarification

Sun, 2012-03-25 17:52
Kanorgon