>
>
> I've never looked at this statistics before, so I'm a bit surprised
>
So am I !
> - I was expecting a larger discrepancy between Wilson B and average B at
> low resolution. Although this is probably because PHENIX uses Peter Zwart's
> likelihood-based Wilson B estimation (Peter - what's the reference?), which
> is supposed to be better.
>
The stability at lower resolutions is indeed due to the likelihood based
method that utilizes a reference curve ('standard wilson plot') as obtained
from the PDB.
The original idea came from Sasha Popov & Gleb Bourenkov, both the reference
curve idea as well as the likelihood target. It is in fact the same target
as used for refinement of anisotropic scale during refinement with the
important note that sigmaA = 0.
This is where the likelihood based wilson scaling comes from:
A.N. Popov and G.P. Bourenkov "Choice of data-collection parameters based on
statistic modeling" Acta Crystallogr. (2003). D59, 1145-1153
These are my notes:
http://www.ccp4.ac.uk/newsletters/newsletter42/articles/CCP4_2005_PHZ_RWGK_PDA.doc
HTH
Peter
>
> Pavel.
>
>
>
> On 7/1/10 12:52 AM, Dirk Kostrewa wrote:
>
> Dear Murugan,
>
> at higher resolution, the Wilson plot captures mainly the contribution of
> atoms with lower B-factors which leads to a systematic underestimation of
> the true B-factor distribution. Accordingly, the average B-factor of refined
> structures tend to be higher then the Wilson B-factor, at least in my
> experience. In your case, it is the other way around. One possible problem
> could be, apart from the fit of the Wilson plot as James Holton suggested,
> that you have reflections at very low resolution with underestimated
> intensities due to cut overloads or measurement in the half-shadow of the
> beamstop. This would result in a too low overall B-factor for the model in
> order to try to fit the usually stronger low resolution reflections at the
> cost of the weaker high resolution data. One quick check of this hypothesis
> is to cut the low resolution at, say, 10 A instead of 50 A and run a
> test-refinement. If this results in more realistic model B-factors, you
> should have a closer look at the low resolution data and exclude the
> ill-measured ones.
>
> Best regards,
>
> Dirk.
>
> Am 30.06.10 19:31, schrieb Vandu Murugan:
>
> Dear all,
> If one could find a difference of more than 15 between Wilson B
> factor of the data ( 55) and Mean B factor of the structure, (30) what could
> be the possible reasons? I am seeing it in my structure. Could someone
> tell me why it could be?? Thanks in advance.
>
> Yours faithfully,
> Murugan
>
>
>
--
-----------------------------------------------------------------
P.H. Zwart
Research Scientist
Berkeley Center for Structural Biology
Lawrence Berkeley National Laboratories
1 Cyclotron Road, Berkeley, CA-94703, USA
Cell: 510 289 9246
BCSB: http://bcsb.als.lbl.gov
PHENIX: http://www.phenix-online.org
SASTBX: http://sastbx.als.lbl.gov
-----------------------------------------------------------------
