Hi Chung-ke,
It is quite confusing with the huge spectrum of relaxation dispersion
models that can be found in the literature covering all different
relaxation dispersion conditions. However you should only use models
when you know that they apply to your system. For the Ishima and
Torchia model (http://wiki.nmr-relax.com/IT99), this has the condition
that pA >> pB, which is quite different to the Carver and Richards
model (http://wiki.nmr-relax.com/CR72) or Nikolai Skrynnikov's
expanded numerical model
(http://wiki.nmr-relax.com/NS_CPMG_2-site_expanded). You really need
to read the primary reference for any model you use. For example in
the Ishima and Torchia paper, you will find the text:
"Equation 4 predicts values of Rex that agree with those calculated
using Equation 3, within 15%, for all values of δωτex, when p1/p2 <
0.15."
Equation 3 is the CR72 model and equation 4 is the IT99 model.
You should note that the IT99 model is what is known in the field of
mathematical modelling as a 'flexible model'. It is more able to
adapt and fit to the data more than the other models. However that
does not mean that it is better. But it will be chosen more than
non-flexible models using Akaike's Information Criterion (AIC) model
selection. There are more advanced techniques in the field of model
selection which deal with such flexible models, for example ICOMP (the
Information Complexity Criterion). But these are not implemented in
relax for now. For an introduction to model selection in NMR, see my
model-free model selection paper at
http://dx.doi.org/10.1023/A:1021902006114. There are some good
textbook references in there which explain all of this, and why ICOMP
would probably not select the IT99 model.
However, I would guess that your aim is not to study model selection
or the IT99 model results, but just to perform a dispersion analysis.
The IT99 paper quote above also indicates that this is a very
approximate model, and is probably not what you are looking for. If
this is the case, don't use the IT99 model! Note that the model
selection is provided mainly to see if the dispersion model fits the
data better than the 'No Rex' model - i.e. it is used to judge if the
dispersion process is statistically significant. You cannot compare
the different dispersion models using model selection as the domain of
validity of most models do not match, they are for different
dispersion conditions, and scientific logic always trumps model
selection.
Anyway, a general rule of thumb is that if you are not sure that a
model is suited for your system, then it probably isn't. You should
also take note of Troels' suggestion of Andy Baldwin's recent model
(http://wiki.nmr-relax.com/B14) which can replace the CR72 model. If
it is all too confusing, then the defaults of simply the 'R2eff', 'No
Rex', 'CR72', and 'NS CPMG 2-site expanded' models should be used.
And maybe you should set the NUMERIC_ONLY value to True so that you
only compare the 'No Rex' and numeric CPMG model. I hope this helps.
Regards,
Edward
On 15 September 2014 04:56, Chung-ke Chang <chungke@xxxxxxxxxxxxxxxxxx>
wrote:
Dear all,
I’ve been tearing my hair out trying to figure out how to choose
relaxation dispersion models for a particular protein. I’ve ran into a
case where many residues are fit to the ‘IT99’ model, but did not fit to
the ’NS CPMG 2-site expanded’ model (relax doesn’t even bother to show the
Chi2 and AIC in the log file). If I understand correctly, the ’NS CPMG
2-site expanded’ model is an exact numerical solution based on the
original Bloch-McConnell equations, whereas ‘IT99’ is an analytical
approximation. I would expect that whatever can be fit with the ‘IT99’
model should also be able to fit to the ’NS CPMG 2-site expanded’ model
since they seem to share the same restrictions (pA > pB). Or am I missing
something?
Cheers,
Chung-ke
_______________________________________________
relax (http://www.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: Question on choosing relaxation dispersion models for proteins