Dear Andreas and all,
refining multiple models together would sound like the most
straightforward solution in a case like that. But (please correct me
if I am wrong), generally, such trick can only work in P1:
obviously, any symmetry operation (but lattice translation) applied to
the shifted model will generate a wrong structure...
Furthermore, is there a way to do an "ensemble refinement " in phenix
(or refmac)? In order to do that correctly one must be able to switch
off all interactions between individual models, while keeping
interactions within each model. Apparently, it can be done in CNS,
although I could not find an example of an input file - can anyone
share an example ?
and, by the way, what stops refmac or phenix developers to account for
lattice translation defects during refinement?
Peter
On 7 Jan 2009, at 11:35, Andreas Förster wrote:
Hey Stephen,
how about simply putting three models, separated by 16A, into your
original unit cell and refining them together. You'd have to guess
their fractional occupancies from the heights of the Patterson peaks
(and make them add up to one, obviously).
This approach (in contrast to correcting the intensities) was the
consensus when I talked to Peter Zwart and Tom Terwilliger about a
similar problem. As my problem still hasn't been solved (probably
because of flawed data processing due to overlapping spots), I'd be
very keen on hearing about your progress.
http://scripts.iucr.org/cgi-bin/paper?S0907444908016648
describes both approaches to lattice-translocated crystals
(correcting intensities and refining several models concurrently).
http://scripts.iucr.org/cgi-bin/paper?ba5111
is the best review on the subject.
All the best.
Andreas
Stephen Hare wrote:
Dear All,
We are currently working on a structure of apparent P21 symmetry
which has been solved by molecular replacement. The data are to
2.7Å but the Rfree will not drop below 30%. The density is clear
for the model we have, however there is extra density that suggests
a shift of the structure by 16Å in either direction - resulting in
three possible overlapping positions for the structure. We assume
this is the result of twinning.
The unit cell dimensions are 102.7Å, 83.0Å, 115.3Å, 90°, 101.8°,
90°. Examining the data with phenix.xtriage also suggests pseudo
translational symmetry with a separation of 16Å. A Patterson peak
at 0.097, 0.000, -0.096 is approximately 30% of the origin peak,
while a second peak of double the translation at 0.192, 0.000,
0.195 is 7% of the origin peak. The structure contains a dimer of
dimers with an NCS 2 fold axis almost perpendicular to the
crystallographic 2 fold. This NCS axis almost coincides with the
diagonal between the A and C axes. A twin axis along the A C
diagonal (l,-k,h) could explain the observed extra density, however
this is not possible because A and C are different lengths.
As a result of the NCS axis running almost perpendicular to the
observed P21 axis, it is possible to merge the reflections in a
larger orthorhombic unit cell - dimensions 137.1Å, 83.3Å, 169.8Å
although here the Rmerge is higher and it is not possible to get a
molecular replacement solution.
Is it possible to define the (l,-k,h) twinning operator in our
original unit cell? or have we missed the actual unit cell?
Or....something else?
Steve
Stephen Hare PhD
Post doctoral research associate
Jefferiss Research Laboratories
Wright-Fleming Institute
Division of Medicine
Imperial College London
Norfolk Place
London W2 1PG
UK
Phone: +44 (0) 20 7594 3908 Fax: +44 (0) 20 7594 3906
--
Andreas Förster, Research Associate
Paul Freemont & Xiaodong Zhang Labs
Department of Biochemistry, Imperial College London
