Hi Wim,
I think that the main question I would ask is how you calculate the
Wilson B factor with 3.0-3.5 A data ...
We have done a rather big study of about 12,000 structures and below
2.5 A, how you calculate the B can result to
major discrepancies. There is hardly a line to get the gradient of. If
you use some scaling to an ideal plot like BEST or ARP/wARP
(same algorithm there) the minimizer and 'spikes' can have a huge
influence.
btw, have you optimized in both cases the Bfactor restraints weight ?
I wonder if that would change things (it will change the FreeR ...)
See Ian Tickle's article
http://journals.iucr.org/d/issues/2007/12/00/gx5119/index.html
I would also expect to use tighter geometry in this resolution btw ...!
(don't be fooled if the current rmsd gives the lower Rfree: if you
change B weight this will very likely change!)
btw, why not use TLS ? For one thing you will get better FreeR and
hopefully better B factor refinement ...!
A.
PS If you want to try things overnight I tend to do something like this:
#!/bin/csh -f
#
echo "xweight bweight rfactor rfree rmszbonds" > stats.log
foreach xweight ( 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.20 0.25 0.30)
foreach bweight ( 0.1 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.5 3.0 )
#
####
#Put you REFMAC script here
# As you can eg obtain it from 'Run and View Command file
# make sure it has these two lines
weight MATRIX ${xweight}
temp ${bweight} 2.0 3.0 2.0 3.0
#
###
echo -n ${xweight} ${bweight} >> stats.log
#The line below will grep the right thing if you do 20 cycles of
refinement 10 TLS, 10 normal in my case
awk '(NF=11){if ($1==20) print $2, $3, $8}' refmac_${xweight}_$
{bweight}.log >> stats.log
end
end
#
On Dec 3, 2008, at 16:59, Wim Burmeister wrote:
I should have given the precision that the problem remains
unaffected by a change of the resolution range (even if I use for
example only 4.5 to 3 A resolution).
I am not using TLS and the data are quite isotropic. Rcryst values
are as expected for such a structure.
Anopther 3.5 A dataset does not show the problem (right column).
Wilson plot B-factor [Å2]
66
43
Refinement
Rcryst
0.186 (0.259)
0.190 (0.239)
Rfree
0.268 (0.408)
0.256 (0.278)
Rms deviations from ideal bondlengths (Å)
0.020
0.018
Rms deviations from ideal bond angles (°)
2.0
1.9
Average B-factor [Å2]
39
46
Values for the highest resolution bin are given in brackets.
Cheers
Wim
Wim Burmeister a écrit :
Dear all,
I have a 3 A structure refined with REFMAC which gives consistently
average atomic B-factors of 40 A2, whereas the B factor from a
Wilson plot is about 60 A2. Is there any explanation for such a
discrepancy?
There are no obvious problems:
No twinning, spacegroup P21 with two molecules in the asu, no
proper ncs symmetry. No pathologic Wilson plot, complete and
redundant dataset (although collected on several crystals with
serious problems due to radiation damage).
Interestingly, the Wilson plot of the Fcalc values is about 60 A2
as for Fobs in the output dataset.
Yours
Wim
--
*******************************************************************************
Wim Burmeister
Professeur, Membre de l'Institut Universitaire de France
Unit of Virus Host Cell Interactions (UVHCI) UMR5233 UJF-EMBL-CNRS
6 rue Jules Horowitz
B.P. 181, F-38042 Grenoble Cedex 9 FRANCE
E-mail: [EMAIL PROTECTED]
Tel: +33 (0) 476 20 72 82 Fax: +33 (0) 476 20 94 00
http://www.uvhci.fr
*******************************************************************************
