*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
Hi,
Of course I agree, individual ADPs are/should be enough/better than TLS at
this resolution*. However, I merely suggest to test for a pronounced global
disorder - if one is present, anisotropic ADPs may not be enough to cope
with it, even at high resolution - and it also may cause macroscopic effects
such as odd diffraction, etc. - which in turn can translate to
less-than-perfect data. I am not sure what restraints are being applied to
individual ADPs in this particular case - it is conceivable that restrained
ADPs may not be sufficient to describe a global displacement of the sort
that is modeled by TLS. Which is why I originally asked if the model might
be over-restrained.
Artem
* In theory, addition of extra parameters should result in an improvement of
R-factors, even if these parameters are strongly redundant with the 3D
principal components of ADPs, yes ? Of course the 'improvement' will be
quite small.
----- Original Message -----
From: "Ethan Merritt" <[EMAIL PROTECTED]>
To: "ccp4bb" <[email protected]>
Sent: Tuesday, January 31, 2006 3:00 PM
Subject: Re: [ccp4bb]: R-factor not dropping
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
On Tuesday 31 January 2006 11:27 am, Artem Evdokimov wrote:
Have you refined overall anisotropicity in SHELX or TLS in Refmac ?
I'm a great fan of TLS, but for a 0.9A resolution anisotropic refinement
I have to point out that both the overall anisotropy and any TLS
components to the ADPs should already be accounted for nicely
in the individual atomic U^{ij} terms. Go ahead and do a TLS
analysis if you are interested in the dynamic properties of your
protein, but don't expect it to improve the R factors of a 0.9A
refinement [*].
[*] This is getting off topic, but here goes...
In principle, including an explicit TLS model could add something
even at very high resolution. This is because the 20-parameter
TLS model describes electron density distributions that are
"banana-shaped". To the extent that the real electrons in your
actual crystal are distributed like that, a complex TLS model could
be a better description than the standard U^{ij} anisotropic ADP
description.
However in practice you can't test this, because the way TLS
refinement is implemented in existing programs is to use it
to generate approximate anisotropic ADPs for the individual
atoms. That brings you right back to the same trivariate
Gaussian description of atomic centers used by standard aniso
refinement, and therefore you will lose the potential benefit
of modeling any higher-order anisotropy.
--
Ethan A Merritt
Biomolecular Structure Center
University of Washington, Seattle WA
