Dear All,
many thanks for all the replies. I have gone with the most
suggested option which was to loosen the sigmas in the dictionary file:
this has worked very well. Other options were to refine in SHELXL
(although George Shedrick cautioned against this - see below) or
phenix.refine. Many thanks to all who replied, cheers, Matt.
Matthew BOWLER wrote:
Dear All,
I have looked everywhere for answers on this one with no luck...
I have several structures at reasonable resolution 1.7 to 1.1A. I
have performed unrestrained refinement on the model which is fine for
the core protein and ligands but the surface atoms start flying off a
bit. So what I would like to do is unrestrained refinement only on
the ligands and leave the protein restraints in place. I have a
feeling that if I cannot do it in Refmac I can in phenix but I cannot
find anything out about that either. All help greatly appreciated,
yours, Matt.
It would be possible with a bit of effort to apply such mixed restraints
using SHELXL. However I'm not convinced that it is such a good idea, at
the resolutions you are working at the geometry of the core regions will
be primarily determined by the X-ray data anyway. Perhaps what we
really need are B-value dependent distance and other restraints, so they
are tighter if the B-values are higher and vice versa.
It would be easy to edit the ligand restraint dictionary & make all the
sigmas very big (assuming the input CIF routines don't perform sanity
checks on the values!): that would make it unrestrained to all intents &
purposes. But when you say that unrestrained refinement is 'fine' what
RMSD's are you getting for the geometry? Typically at around 1.5 Ang
resolution you get an RMSD(bonds) of around 0.2 Ang for unrestrained
refinement which is obviously quite big compared with the average SU of
around 0.02, and for restrained refinement you expect an RMSD even
lower, around 0.01 at that resolution, because of the need to avoid
over-fitting. Admittedly those are averages over the whole structure so
it will be less in the core & more on the surface, but I would be
surprised that even for the core atoms you would get sensible RMSDs.
We did this by naming atoms CCX or OHX and
not defining the bond lengths or by giving v v v low
weighting to the bond length energy constant
in the definition of that bond. We did this too look
at a specific carboxyl binding calcium.
SHELX is also a good way to go.
Same question bothered me for a while.. if one could define a mask or
a subset of atoms as PDB and even release restrains globally.. if you
figure out how to do it in refmac, please let me know. The
phenix.refine way is the following, as given in the documentation..
Removing selected geometry restraints
In the example below:
% phenix.refine data.hkl model.pdb remove_restraints_selections.params
where remove_restraints_selections.params contains:
refinement {
geometry_restraints.remove {
angles = chain B
dihedrals = name CA
chiralities = all
planarities = None
}
}
you can increase the sigma's of the restraints 10-fold. The effect will be
quite similar to unrestraint refinement. (and delete those planar and
chiral restraints)
--
Matthew Bowler
Macromolecular Crystallography Group
European Synchrotron Radiation Facility
B.P. 220, 6 rue Jules Horowitz
F-38043 GRENOBLE CEDEX
FRANCE
=============================================================================
Tel: +33 (0) 4.76.88.29.28
Fax: +33 (0) 4.76.88.29.04
http://www.esrf.fr/UsersAndScience/Experiments/MX/About_our_beamlines/ID14-2/
=============================================================================
