DETWIN

["Leslie A." <andrew@mrc-lmb.cam.ac.uk>]

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***



Bryan Wang was asking about the correlation coefficient output by
DETWIN:

*****************************************************************************

*** Twin Fraction of 9.6%

 Correlation coefficient (on intensities) for input data 0.582
 Correlation coefficient after detwinning 0.425

 Number of reflections written to output file 30340
 Number omitted because the twin mate was not observed 7

  Correlation coefficient as a function of resolution. 
    Dmin Number CC before CC after
                   detwinning detwinning
    7.91 1135 0.395 0.203
    5.82 1819 0.275 0.070
    4.83 2299 0.273 0.067
    4.21 2704 0.277 0.071
    3.78 3033 0.296 0.092
    3.46 3320 0.483 0.305
    3.21 3636 0.635 0.491
    3.01 3933 0.743 0.632
    2.84 4101 0.390 0.197
    2.70 4360 0.294 0.091


*** Twin Fraction of 24%

 Correlation coefficient (on intensities) for input data 0.582
 Correlation coefficient after detwinning 0.012

 Number of reflections written to output file 30340
 Number omitted because the twin mate was not observed 7

  Correlation coefficient as a function of resolution. 
    Dmin Number CC before CC after
                   detwinning detwinning
    7.91 1135 0.395 -.231
    5.82 1819 0.275 -.355
    4.83 2299 0.273 -.358
    4.21 2704 0.277 -.354
    3.78 3033 0.296 -.335
    3.46 3320 0.483 -.126
    3.21 3636 0.635 0.095
    3.01 3933 0.743 0.294
    2.84 4101 0.390 -.237
    2.70 4360 0.294 -.337


Are these results consistent with the twin fraction estimate of 9.6% being
correct?

*****************************************************************************


The answer to the question (is 9.6% twinning correct?) is YES. 

It is important to look at the correlation coefficient as a function
of resolution rather than the overall value. The overall correlation
coefficient will always be significantly greater than zero (eg
0.4-0.5) even for data that are not twinned at all. This simply
reflects the fact that the data gets weaker (on average) at higher
resolution. In this case, the twin fraction of 9.6% is giving very
small correlation coefficients in each resolution bin (except 3-3.4A)
after detwinning, which is a good sign, while the twin fraction of 24%
is giving large -ve correlations, which is definately a bad sign.


I do not, however, have an explanation of why the correlation
coefficient peaks at 3A (after detwinning using 9.6% twin fraction). I
do not know how frequent an observation this is (has anyone else seen
the same effect ?). Is there any non-crystallographic symmetry in your
system ?


               Andrew Leslie