Hi,
For reference, in the future can I ask you not to attach files to
public mailing list messages. Cheers! Attachments cause a large
strain on the open source infrastructure behind relax. If you need to
attach files, then this can be performed using the relax trackers
(bugs, tasks, support request, patch, etc) at
http://gna.org/projects/relax.
For the results, note that what you have calculated is only a small
fraction of a full model-free analysis. Firstly, how did you
determine the initial diffusion tensor?
Your next step will be to optimise the diffusion tensor together with
all model-free parameters, i.e. the 'global' model. Note that the
starting diffusion tensor will be different to the final diffusion
tensor. So then you need to repeat everything using the final
optimised diffusion tensor as the initial diffusion tensor. Using a
different diffusion tensor at the start (i.e. the final optimised
tensor) will give you different final results. After repeating, you
can then compare the optimised diffusion tensors again and see that
they are different, and that the model-free results are also
different. You have to repeat this up to 20 times (sometimes much
less, other times much more) until the chi-squared value is identical
between two rounds (to the last decimal place), the model-free models
and parameters are identical, and the diffusion tensor parameters are
identical. Please see my 2007 review
(http://dx.doi.org/10.1039/b702202f), and 2008b paper
(http://dx.doi.org/10.1007/s10858-007-9213-3) for details about this
iterative procedure. There are other useful details in my other
papers (see http://www.nmr-relax.com/refs.html). And the model-free
chapter of the relax manual also goes into a lot of detail about this
iterative procedure (http://www.nmr-relax.com/manual/).
You will then need to complete this iterative procedure separately for
the spherical, oblate and prolate spheroidal, and elliptical diffusion
tensors. You then need to perform model-selection between these
global models (AIC is useful here as these are non-nested models and
hence F-tests are not applicable). This is all part of current and
past model-free protocols.
Also note that there can be a circular pattern in your iterative
optimisation. This combined optimisation/model selection problem can
spin in a perpetual loop around the universal solution (the solution
in the universal set, which is the combination of all global models of
the diffusion tensor plus all model-free models and the optimisation
spaces of each global model). For more details, see my 2007 paper.
So you will need to catch this circular pattern and terminate
optimisation when it is reached. For reference, the
dauvergne_protocol auto-analysis built into relax performs such checks
(auto_analysis/dauvergne_protocol.py).
Only once all of this is completed will you be ready to use the
value.write and grace.write user functions for data visualisation.
Until then there is a lot of literature to catch up on ;)
Regards,
Edward
On 8 January 2013 13:57, <mengjun.xue@xxxxxxxxxxxxxxxxxxxx> wrote:
Dear Dr. Edward d'Auvergne,
I have tried to run mf_multimodel.py under relax-2.1.2 for analysis of
single field data (demo data ubq), it seems it work well. I also run
modsel.py, the model M5 is good for most residues. I would like to ask you
about the output file, the model free parameters calculated (for example,
M5) are dispersed in the text of output file (resultS.bz2, or log file), how
to get the model free parameters from these output files, so that the
parameters can be input to other software,for exapmple, origin or sigmaplot.
Attached please find the log file or results.bz2 for M5. Thank you.
With best regards,
Mengjun Xue
Re: How to get model free parameters from output files