[phenixbb] Twinning Refinement

Fri Mar 26 08:28:38 PDT 2010

Dear Joe,

It is not uncommon for proteins to crystallize in different crystal forms when in complex with different substrates or inhibitors.  Trust that the absurdly high R factors (38% in the lowest resolution shell!?!?) despite the great lengths you went through with error models, etc. and the fact that scalepack is trying to throw out 20% of your data is a hint that you may be in the wrong spacegroup.  Also, most of your "systematic absences" have an I/sig >2.5 and so probably are not absent. Were the predictions falling on spots during integration?  This may provide another clue.

I think you need to go back to HKL and reprocess pretending that you don't know that the crystals are "well characterized".  You may have a new crystal form.

good luck,
Eric

__________________________
Eric Larson, PhD
MSGPP Consortium
Department of Biochemistry
Box 357742
University of Washington
Seattle, WA 98195

> Date: Fri, 26 Mar 2010 05:13:07 +0000
> From: Joseph Brock <joeylives2ride at hotmail.com>
> To: <phenixbb at phenix-online.org>
> Subject: [phenixbb] Twinning Refinement
> Message-ID: <SNT120-W557FCF7E111C0FEA167769B230 at phx.gbl>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Dear Phenix Users,
>
> I'm currently using version:
>   Version: 1.4
>   Release tag: 3
>   cctbx tag: 2008_12_07_1353
>   Platform: mac-intel-osx osx-10.5.6
>
> I am working on characterising substrate/inhibitor interactions of a protein, the structure has been solved many times so its unit cell dimensions, contents (1 molecule per ASU) and space group (P3121) are well characterized.
>
> However, I have recently encountered several crystals that appear exhibit merohedral twinning, and am having trouble converging the R-factors to reasonable levels. The details of one such data set is described below:
>
> Diffraction to ~1.9A was recorded using a high and low resolution pass of 180? each. When processed with HKL2000 separately or together, Scalepack wanted to reject huge amounts of my data (~20%). I was forced to increase the error scale factor to 2 and  adjust my error model to the very high R-factors of the low-resolution shells. The summary table from my final round of scalepack is:
> Shell Lower Upper Average      Average     Norm. Linear Square
> limit    Angstrom       I   error   stat. Chi**2  R-fac  R-fac
>      50.00   4.09  7648.3  1498.7 182.7 1.030  0.376  0.424
>       4.09   3.25  3885.1   568.9   64.1  1.410  0.416  0.480
>       3.25   2.84  1609.7   225.0   26.1  1.227  0.451  0.517
>       2.84   2.58   807.9    97.1     25.3  1.958  0.439  0.561
>       2.58   2.39   496.6    59.7     22.0  1.060  0.257  0.268
>       2.39   2.25   335.5    53.0     18.1  0.717  0.275  0.238
>       2.25   2.14   221.7    19.0     17.4  2.389  0.354  0.298
>       2.14   2.05   143.1    19.5     17.5  1.537  0.459  0.351
>       2.05   1.97    87.0    20.2      17.1  1.079  0.663  0.480
>       1.97   1.90    54.6    20.3      16.7  1.073  0.990  0.696
>  All reflections   1588.3   269.2    41.9  1.350  0.400  0.444
>
> In addition, many systematic absences appear to be present:
>
>     Intensities of systematic absences
>      h   k   l  Intensity     Sigma   I/Sigma
>
>      0   0   5     177.7      56.7       3.1
>      0   0   7     255.5      70.6       3.6
>      0   0   8     264.8      73.2       3.6
>      0   0  10     595.0     164.2       3.6
>      0   0  11    2343.5    1289.9       1.8
>      0   0  13     321.3     125.4       2.6
>      0   0  14      66.5      21.8       3.1
>      0   0  16     238.0      76.3       3.1
>      0   0  17     394.3     109.1       3.6
>      0   0  19      13.0       7.1       1.8
>      0   0  20     650.1     179.5       3.6
>      0   0  22     199.6      55.6       3.6
>      0   0  23     953.9     372.0       2.6
>      0   0  25     192.2      62.2       3.1
>      0   0  26    1038.7     405.0       2.6
>      0   0  28      96.3      28.9       3.3
>      0   0  29    2303.0     897.0       2.6
>      0   0  31      12.0      11.5       1.0
>      0   0  32    1622.4     516.5       3.1
>      0   0  34      11.5      13.8       0.8
>      0   0  35      22.8      14.8       1.5
>      0   0  37      27.9      18.0       1.5
>      0   0  38     154.3      45.6       3.4
>      0   0  40      55.8      27.8       2.0
>      0   0  41      35.5      23.7       1.5
>      0   0  43     -25.3      34.4      -0.7
>      0   0  44      67.7      32.1       2.1
>
>
>
> The data has an overall completeness of 99.7%.
>
> When I run the data trough xtriage, it tells me:
>
> Statistics depending on twin laws
> -----------------------------------------------------------------
> | Operator | type | R obs. | Britton alpha | H alpha | ML alpha |
> -----------------------------------------------------------------
> | -h,-k,l  |   M  | 0.307  | 0.301         | 0.256   | 0.022    |
> -----------------------------------------------------------------
>
> Patterson analyses
>  - Largest peak height   : 5.288
>   (corresponding p value : 0.86552)
>
>
> The largest off-origin peak in the Patterson function is 5.29% of the
> height of the origin peak. No significant pseudotranslation is detected.
>
> The results of the L-test indicate that the intensity statistics
> are significantly different than is expected from good to reasonable,
> untwinned data.
> As there are twin laws possible given the crystal symmetry, twinning could
> be the reason for the departure of the intensity statistics from normality.
> It might be worthwhile carrying out refinement with a twin specific target function.
>
> I also estimated the twin fraction with the following server:
>
> http://nihserver.mbi.ucla.edu/Twinning/
>
> And found it to about 20% (using data from 6-3A).
>
> After phasing the data with rigid body refinement in Refmac with intensity based twin refinement (brings both R-factors to ~36%), I have tried doing a twin refinement with Phenix:
>
> %phenix.refine model.pdb data.mtz twin_law="-h,-k,l" strategy=rigid_body+individual_sites+group_adp
>
>
> The R_factors consistently diverge to the following values:
>
> Final R-work = 0.3392, R-free = 0.4360
>
> Simulated annealing/TLS/ordered solvent/messing with wxc/wxu weights does not seem to improve the situation. I have also tried this with and without prior detwinning with the CCP4 program "Detwin", with similar results.
>
> Reindexing to -h,-k,l gives similar R-factors after rigid body refinement but after refinement with Phenix, things get even worse:
> Final R-work = 0.3919, R-free = 0.5059
>
> I was surprised to see that the fit to the map using both indexing conventions are comparable even though this should effectively reverse the hand of the structure. I assume this is because twin refinement is basically doing the job of "partial reindexing" on those parts of the lattice which are twinned?
>
> I have been working on this for about a week now and can't for the life of me find a way forward. If anyone can offer any suggestions It would greatly be appreciated. Im slowly beginning to lose my sanity....
>
> BTW I don't think my problems stem from misindexing, although the beam was wondering all over the place during this block of data collection, I have previously solved another structure at the same d and xbeam ybeam values without any trouble.
>
> Sorry about the v.long post. Thought it would be better to give too much information rather than not enough =)
>
> Many thanks for all your suggestions in advance and for everything I
> have learned from this community in the past. Cheers,
>
> Joe
> PhD student.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> _________________________________________________________________
> Get personal with Windows. Download a free gift for your PC.
> http://experience.windows.com
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20100326/1f6df68c/attachment-0001.htm>
>