Sun,
I'm most of the way to one side of this debate: I believe that it is
not possible to emerge fully from model bias without avoiding it in
the first place with experimental phases. I may be overly
pessimistic, but have considerable experience supporting at least
skepticism.
My interpretation of the experimental result you describe is that the
covariances among the parts of the structure you left in place and
those side chains you omitted is so strong and extensive that you'll
never see the correct density coming back upon refinement, because
other parts of the structure are ever so slightly off their true mean
positions to compensate for the (evidently false) positions of the
residues you omitted. Bill's suggestion that you actually refine the
structure using simulated annealing without the omitted residues is
an improvement over what you did, but it will require many cycles to
get a much better approximation, and there is really no way to be
sure when you can be confident. Starting the entire refinement over
is a more aggressive strategy. If you decide to try this, you should
examine the projection of the residue by residue real-space
correlation coefficients across the entire sequence to ensure that
you have only one population of values and delete all residues that
comprise any population that has a distinctly different real-space
correlation coefficient, building them back into the structure as it
refines. That is, you should ensure that you don't begin refining any
residues at the very beginning for which there is evidence that they
might be different from their positions in your molecular replacement
model.
Charlie
On Jul 26, 2008, at 2:12 PM, William G. Scott wrote:
Hi Sun:
It might be worth doing a simulated annealing omit refinement in
phenix or CNS, with the residues in question omitted. CNS also
allows you to make a composite-omit map. I haven't seen that in
phenix yet but presumably it is doable.
Bill
William G. Scott
Contact info:
http://chemistry.ucsc.edu/~wgscott/
On Jul 25, 2008, at 10:53 PM, Sun Tang wrote:
Hello Everyone,
I have a question about getting rid of model bias in refinement
with refmac. I solved the structure with molecular replacement.
After final refinement of the structure, I found out some key
amino acids in the structure and wanted to make sure their
conformations are correct. I omitted these amino acids (by setting
occupancy to zero) and refined the structure. I manually fit the
amino acids into the density and refined the structure again. I
found these amino acids return to the precious conformations even
though the conformations I fit were different. Should I omit these
amino acids from the beginning of the refinement? What is the best
way to get rid of the model bias? Your suggestions are greatly
appreciated!
Best,
Sun
**********UNCrystallographers NOTE new website url******************
[EMAIL PROTECTED]
http://xtal.med.unc.edu/CARTER/Welcome.html
Department of Biochemistry and Biophysics CB 7260
UNC Chapel Hill, Chapel Hill, NC 27599-7260
Tel: 919 966-3263
FAX 919 966-2852
