Hi,
This problem you have encountered is that of artificial chemical
exchange. This is quite well documented by Tjandra et al., 1995. The
instability has been compounded by the use of only single field
strength data as the physics theory cannot differentiate between Rex
(which scales quadratically with field strength assuming fast
exchange) and the global correlation times (which are field strength
invariant). I would recommend also reading the following article:
d'Auvergne E. J., Gooley P. R. (2007). Set theory formulation of the
model-free problem and the diffusion seeded model-free paradigm. Mol.
Biosyst., 3(7), 483-494.
This article is linked from here: http://www.nmr-relax.com/refs.html.
In this paper I fully describe all of the issues you are seeing here,
covering all details and possible solutions. I hope this helps. If
you have any questions from this paper, please don't hesitate to ask.
Regards,
Edward
On Thu, Jan 8, 2009 at 4:43 PM, Pierre-Yves Savard
<Pierre-Yves.Savard@xxxxxxxxxxxxx> wrote:
Hello,
I have some trouble using relax with my data.
I have data at 600 MHz only so I use models m0,m1,m2,m3,m9 with the
classic approach of starting with an initial estimate of the Tm, with a
modified versiion of the script mf_multimodel.py. which includes model
selection and diffusion tensor optimization.
The first problem is that it doesn't converge. However, the second
problem is more important in that, following the different iterations,
the Tm and S2 go down and the Rex go up to values approaching the R2.
After 35 iterations, the Tm is less than half that estimated with the
R2/R1 ratio (using quadric or tensor2) and almost all residues are
fitted with model m3... This problem starts at the second iteration.
I used this approach with success using a sample of the same protein
(136 residues) with a slightly different concentration. Using the data
from this sample (which are very close to the problematic dataset), the
converged tm was very close to the estimated tm based on the R2/R1
ratio... Moreover, S2 were much higher, with a mean value of 0.8, with
only a few Rex...
Sample one (good results): Estimated Tm = 10.1 ns, converged Tm = 10.1,
S2 = 0.88, 8 residues with Rex (mean of 2 s-1).
Sample two (problematic): Estimated Tm = 10.09 ns, Tm after 35 iteration
= 4.30, S2 = 0.48, 102 residues with Rex (mean of 10.1 s-1).
Any idea why the second sample gives rise to such annoying results ?
Thanks a lot,
Pierre-Yves
--
Pierre-Yves Savard
Professionnel de recherche
Département de biochime et de microbiologie
-------------------------------------------
Pavillon Charles-Eugène Marchand, local3252
1030 avenue de la médecine
Université Laval, Québec, Qc, G1V 0A6
Tél: 418.656.2131 #4530 Fax: 418.656.7176
Pierre-Yves.Savard@xxxxxxxxxxxxx
-------------------------------------------
_______________________________________________
relax (http://nmr-relax.com)
This is the relax-users mailing list
relax-users@xxxxxxx
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-users
Re: Very low Tm in model-free analysis