Hi Christina,
Welcome to the relax mailing lists! As Troels mentioned, I have been
on holidays so can only now reply. For details, please see below:
On 3 June 2015 at 14:32, Christina Möller <c.moeller@xxxxxxxxxxxxx> wrote:
Dear Edward and relax users,
I successfully performed the dauvergne_protocol.py analysis scripts to
determine the ps - ns dynamics of a 14 kDa protein. The global
correlation time tm is 9.6 ns.
This value does not seem unreasonable.
Since other methods suggested a smaller
global correlation time tm,
Which other methods have you used? You should note the following issues:
- It is well known that MD underestimates the global correlation time
by about half and David Case is actively researching this area.
- Stoke's law and the related hydrodynamic beads model from Garcia de
la Tor are reasonable for estimating tm for isolated molecules.
However with the super high concentration in NMR samples, these tend
to underestimate the tm value by about half as they do not take
micro-viscosity into consideration.
- ORD measurements as well, as the tm changes between the lower
concentration optical spectroscopic sample and the NMR sample by about
a factor of 2 (micro-viscosity effects being a major factor again).
I would like to know whether it is possible
to fix the tm value in all rounds of individual model-free
optimisations?
As Troels pointed out, this is of course possible. You can take, for
example, the sample_scripts/model_free/diff_min.py script. This
already implements a lot of what Troels described for you. Note
however that this needs to be executed iteratively until the global
chi-squared value between iterations is identical. See figure 2 of:
d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR
dynamic models II. A new methodology for the dual optimisation of the
model-free parameters and the Brownian rotational diffusion tensor. J.
Biomol. NMR, 40(2), 121-133. (
http://dx.doi.org/10.1007/s10858-007-9213-3 ).
This global iterative optimisation of the diffusion tensor is
essential, and you'll find it documented in Mandel et al., 1995 as
well as in the papers by the Dasha authors. It will take between 5-20
iterations to properly converge (in rare cases it can be 2 iterations,
in others >50). For more details, you really should read:
How to get model free parameters from output files (
http://thread.gmane.org/gmane.science.nmr.relax.user/1375/focus=1378
).
This requires an initial diffusion tensor estimate. And as I
demonstrated in the above paper - as well as first shown in the
Korzhnev et al., 1999 bacteriorhodopsin fragment paper (
http://dx.doi.org/10.1023/a:1008356809071 ) - if this is too far off,
the global minimum will never be reached. You can however directly
compare the two results using AIC values (not chi-squared values as
the individual model-free models for each residue will be different
and the effects of parsimony will not be taken into account). An
additional thread which might be of interest is:
AIC to select diffusion model (
http://thread.gmane.org/gmane.science.nmr.relax.user/885/focus=891 ).
There are plenty of other relax-users mailing list threads on the
subject which you can search for at:
http://dir.gmane.org/gmane.science.nmr.relax.user
The corresponding chi2 values might then be useful to
evaluate the global correlation times that I obtained by different methods.
Note that you should compare the chi-squared values from the same
program, just to be sure. Also note that the spherical angle and
Euler angle notations in Modelfree4, Dasha, Tensor2 and relax are not
compatible. The problem is that the definitions of these angles are
not documented (except in relax) so if you take a diffusion tensor
from one and input it into another, you will see the angles swing
around wildly with the global iterative diffusion tensor estimate
until it converges to the same tensor but with the different angles
(except when a local minimum is hit). There are 2406 Euler angle
conventions and symmetries for diffusion tensors!
One last thing to note is that relax is extremely flexible in what it
can do. Using specially designed scripts, relax can replicate the
results of Modelfree4, Dasha, Tensor2, or DYNAMICS. One exception is
that relax uses a real optimisation constraint algorithm (the
augmented Lagrangian or method of multipliers,
https://en.wikipedia.org/wiki/Augmented_Lagrangian_method ,
https://gna.org/projects/minfx/ ) which the other model-free softwares
do not, hence there can be cases where relax does not find exactly the
same result as the other softwares.
I hope all this information helps. You should also consider Troels'
suggestion of the dx.map user function to see the diffusion tensor
parameter space (or 3D subsets of it). I used this in figure 6 of:
d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR
dynamic models I. Minimisation algorithms and their performance within
the model-free and Brownian rotational diffusion spaces. J. Biomol.
NMR, 40(2), 107-119. ( http://dx.doi.org/10.1007/s10858-007-9214-2 )
Regards,
Edward
Re: fixed tm values