On Mon, 15 Apr 2019 at 19:18, Boeszoermenyi, Andras
<[email protected]> wrote:
>
> Dear Relax Community,
>
> I am trying to do model free analysis with relax 4.0.3 on a linux workstation
> in script mode.
>
> I have a high resolution (1.1 Angstrom) crystal structure of a 30 kDa protein
> and added protons to the pdb.
> For the model free I have T1, T2, HetNOE data sets at 600 MHz and 750 MHz
> (both spectrometers were Bruker) and all values make sense. The analysis of
> the data fits well, with small standard deviations for the vast majority of
> residues.
>
> I am using the "black-bocks" dauvergne-protocol and all calculations run
> perfectly fine.
>
> However, the Chi-squared test produces ridiculously high chi-squares ~17000
> and then of course the wrong model is selected.
Hi Andras,
This could be due to a number of factors:
- Missing or poor temperature calibration or control (see the
"Temperature control and calibration" section of the "Relaxation
curve-fitting" chapter of the manual, i.e.
http://www.nmr-relax.com/manual/Temperature_control_and_calibration.html
).
- Inadequacy of the single diffusion tensor estimate. Please
compare the local tm model results to the final selected diffusion
model. What is the total chi2 value for the local tm model? Could
there be partial dimerisation, even if it is non-specific?
- Incorrect errors in the R1, R2, and steady-state NOE values.
Did you use relax to obtain this data? And how did you perform the
error analysis? If the errors are too small, the final chi-squared
value will be very high and the model-free models chosen will be
wrong. See the "From spectra to peak intensities for the relaxation
rates" section of the "Relaxation curve-fitting" chapter of the
manual, i.e.
http://www.nmr-relax.com/manual/From_spectra_to_peak_intensities_for_the_relaxation_rates.html
. And, for example, the "NOE script mode - setting the errors"
section of the "Calculating the NOE" chapter of the manual, i.e.
http://www.nmr-relax.com/manual/NOE_script_mode_setting_the_errors.html
.
There are other factors as well. I suggest looking at the relaxation
curves for the R1 and R2 very carefully, including error bars, to see
if there is any issue with the base data. These curves are produced
automatically by the relax auto-analyses and can be visualised in
XMGrace. The steady-state NOE data and parameter plots should also be
carefully checked.
> Did anyone make this experience, and would any one have a suggestion on how
> to fix it?
See above. This has been seen many, many times before. The above 3
reasons are probably the most common ones - with temperature
calibration and control being the number one culprit. You may be able
to find the old threads on the relax users mailing list archive on
this subject: https://sourceforge.net/p/nmr-relax/mailman/nmr-relax-users/
> One more information. I ran the NMR experiments on a perdeuterated and 13C,
> 15N labeled sample. I did not find a way to correct for 13C couplings in the
> model free script. Might that have a role in this, and is there a way to
> correct for that?
This will be an issue. I have never encountered literature on the
subject of dealing with this on the data analysis level. I thought
that this is usually dealt with by minimising the effect during the
experiment. Do you have any references/papers that state otherwise?
I would be interested in reading about it.
I hope the above info helps.
Regards,
Edward
_______________________________________________
nmr-relax-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/nmr-relax-users
