*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
First of all I want to thank everybody, for replying this fast.
SUMMARY of replies:
Q: Refinement Problem in Refmac5 (Problem in MAKE_U_POSITIVE)
I would like to know if it is legitimate to use tls only on parts of the
asymetric unit and/or if there are any other things I should look for.
Answer (Ethan Meritt):
Sure. There is no guarantee that any particular protein molecule, or
piece of a protein molecule, is well-described by a TLS model. It often
is, but not always.
It may also be the case a multi-domain protein that the individual
domains have large TLS components, but the entire protein taken as
a whole is not well described as a single TLS group. In such a case
you can get a dramatic improvement in R and Rfree by breaking the
protein into multiple TLS groups. However, based on the R factors
you report I doubt your structure is an example of this.
Feed your model to the TLSMD server and see what it says about
your individual chains.
http://skuld.bmsc.washington.edu/~tlsmd
Answer (Martyn Winn):
The MAKE_U_POSITIVE problem may be genuine, i.e. some
parts of your structure don't fit the assumptions of
the TLS model. In which case, sure, leave them out.
Or it may simply be a symptom of the refinement
protocol. Starting with fixed constant Bs allows the
TLS full freedom to describe the displacements. But maybe
too much freedom. In which case, having some variation
in the Bs is useful. One protocol is to keep the residual
Bs from one round, and start the TLS parameters from
zero again.
Answer (Jay Painter):
The 'Problem in MAKE_U_POSITIVE' can be a indication that your protein
chain is not well described by a single TLS group. The TLSMD server
should select multiple groups which will help with this problem. It
works best if you also use the TLSMD server to generate input PDB and
TLSIN files for further refinement.
I have also found it necessary to patch REFMAC5 to keep TLS refinement
stable when using a large number of TLS groups for each protein chain.
I'm attaching the current version of my patch against the ccp4-5.0.2
source tree. If you want to use it, you'll need to apply it then
re-compile refmac5 like this:
# cd ccp4-5.0.2
# patch -p1 < ../ccp4-5.0.2_refmac5_tlsmd_fix_v1.patch
Answer (Bryan W. Leopore):
i have tried feeding the model back into
refmac, and things stabilized.
Answer (Valerie Bio):
I once had this type of problem (Bfactor calculated with Truncate was
50 and the average Bfactor from Refmac+TLS was 20) and by removing the
very low resolution in the refinement (between 30 and 20Å), the average
Bfactor rose to 50.
This did not change the Rfactors much.
And now here what helped in the end:
Since I already tried to break the individual chains into several tls
groups according to the flexible domains from a analysis with escet and
this showed some improvement I send the coordinates to the TLSMD server.
Using the output of this server with 2*20 (chains C,D) and 2*15 (chains
A,B) tls groups as input for a refmac run helped a lot. After this just
about 10 AA in chain D caused problems. Tighten the Bfactor restrains in
refmac fixed this problem. I used also the patch for refmac provided by
Jay.
Feeding this refined model back to refmac and setting the tls parameters
to zero was successfull.
The R/Rfree is now 17/22 (before 21/27) and the Density improved.
Switching to simple scaling and/or removing low resolution data didn't
help.
Keeping the residual Bs from one round and to start the tls parameters
from zero didn't help before I used the TLSMD server output and tuned
the Bfactor restraints.
Puzzling to me was that chain D was the troublemaker although this was
the Protomer with the lowest Bfactors and the best electron density.
best regards
Christian
-
Christian Schleberger
Universität Freiburg
Inst. für org. Ch. & Biochemie
