On Wed, Mar 11, 2009 at 2:00 PM, Eldon Ulrich <elu@xxxxxxxxxxxxx> wrote:
Hi,
I guess I am a little confused by the comments on the '.model_fit' tag and
its values. My interpretation of the XML container example is; here are a
list of possible models that the data could be fit to (m(0) to m(9)). For
this particular set of data this model was used. The m(0) to m(9) are equal
to a specific set of parameters that define the model.
In the relax output there is:
<model desc="The model">
'm5'
</model>
<equation desc="The model equation">
'mf_ext'
</equation>
<params desc="The model parameters">
['S2f', 'S2', 'ts']
</params>
The model name and equation can be deduced from the parameters used.
So the spin specific .model_fit tag is more like the 'params' section
of the XML file.
For the NMR-STAR example, the value assigned to the tag '.model_fit' defines
the specific set of parameters that define the model that was fit in
calculating the values reported. For each row in the NMR-STAR table, a
different model may have been fit.
This is true. After model selection, each spin will have a different
model. That brings up another possible tag for the top of the
saveframe - the model selection method used. Note that model
selection is a ~100 year old, intensely studied field of statistics
(http://en.wikipedia.org/wiki/Model_selection) and that different
methods can be used here!
As I understand Art's ModelFree program, the data were fit to all five
models and the model that fit the best was reported. Thus, within a table of
amide order parameters for a protein, one model may have been used for one
residue and another model for another residue.
This is model selection, but the model selection used originally goes
against this ~100 year old field (and ignores it). relax uses by
default something called AIC model selection
(http://en.wikipedia.org/wiki/Akaike_information_criterion). But this
is all true - a different model-free model can be used for the
backbone N and Ca for even the same residue.
I am not an expert in this and am walking on thin ice here,
Your understanding is perfectly correct. It would just be nice to
have a fixed set of values for '.model_fit', which can be different
for each spin, to allow this to be parsable for automatic reading into
relax or other programs.
Regards,
Edward
Re: To all relax developers: presenting relax at the ENC conference and preliminary BMRB NMR-STAR file creation for model-free results.