Hi Peter,
After you have finished checking the data again for twinning, etc.
as suggested by Eleanor,
you might try the following:
1) Refine using the simulated annealing refinement protocols in CNS and
then check R-values and map quality. If you try that, you can also test
a few different starting temperature values and slow cooling/constant
cooling during the annealing procedure. Following this, you can generate
composite omit maps and also try Primer and Switch phasing (look up
RESOLVE manual) to try to reduce model bias and try to improve map
quality which may allow you to trace more of the model.
2) What is the sequence identity of the target compared to the
homologous search model? Have you already changed all the sequence of
the MR model solution to reflect your actual target sequence? This may
help to improve your R-values.
3) Seems that you may have several different homologous models to choose
from. You may re-try MR using different homologs and see which gives you
the best starting R/Rfree and select that and change as many amino acids
as you can to reflect your actual target sequence and then try
refinement again.
4) Use some homology modeling program and try to remove elements of
secondary structure from your MR solution that may potentially not be in
your actual target or may be unstructured (insertions, deletions, etc.)
It is not extremely unusual to get a MR solution using a dimeric search
model instead of the individual monomers. You may also have the R
values given the 40% completeness and also the fact that the search
model is of a homolog.
Hope some of this may help your case although it may still take cycles
of iterative manual building.
Else, since you have 60% of the model missing, you may just have to go
for experimental phasing. Since you have a starting MR solution and thus
initial partial phases, even if you can get one or more low/medium
resolution derivatives on your needle crystals, you can use difference
Fourier methods to supplement the partial MR phases with experimental
phases for a complete structure determination. Of couse, if you are able
to improve your crystals, you should have more robust crystals for heavy
atoms soaks.
Thanks,
Debanu.
Eleanor Dodson wrote:
I guess I would go back to the data.
Could there be twinning?
Is the a pseudo translation vector which means there are
systematically weak zones of data? ( hklview might show something
strange)
Rfree seems rather close to R to me for this resolution - maybe it
should go up a bit?
How many cycles have you done already? PHaser refines without taking
FreeR into account I think..
And is there density for the missing bits?
Eleanor
Peter J Stogios wrote:
Hi,
Long-time reader, first-time writer to ccp4bb.
I'm having a problem with a 3.0 angstrom dataset (2.8 if I push it).
These crystals were thin but very long needles and I could see
diffraction only at the synchrotron. Diffraction spots were very
very small but well defined. High level of radiation damage.
P2(1)2(1)2(1), no large unit cell axes. R-int 7%(14 in highest
shell), I/sigma 14(8), completeness 98%. Nothing weird so far and
the stats actually look good. I was ecstatic these tiny needles
diffracted at all!
I expect 2 chains by Matthew's coefficient and 50% solvent.
Biological unit is a dimer, homologs are dimers, this protein
purifies as a dimer. I'm solving it by molecular replacement, using
Phaser, which will give a refine-able solution only when searching
with dimeric models from homologous proteins. I cannot get solutions
that make sense or give Z-scores higher than 5 when searching for two
chains. But that doesn't matter, I get a solution when I search for
a dimer that is able to refine.
So far so good. Here's where I get stuck: refinement with Refmac
goes well until R/Rfree values of 32/35, but I cannot break this
barrier. In fact, building into positive Fo-Fc peaks results in
R-free getting worse--actually any further refinement at all results
in R-free going up. The model is only 40% complete and has many
missing regions, not only in loops in turns. I just can't improve
the model anymore. I've tried rebuilding in resolve, Arp/warp, and
of course lots of manual building.
I don't think my data is as bad as to restrict my refinement, so I'm
confused as to why I've hit this barrier. Hopefully this description
isn't too vague but I appreciate any help in advance and I can
elaborate if you're willing to help!!
~
Peter J Stogios
Ph.D. candidate, Privé Lab
Dept. of Medical Biophysics, University of Toronto
Toronto Medical Discoveries Tower (TMDT) at MaRS
101 College St., Rm. 4-308
Toronto, Ontario M5G 1L7
e: [EMAIL PROTECTED]
w: http://xtal.uhnres.utoronto.ca/prive
p: (416) 581-8550 ext. 7543
