local scaling

[Phil Evans <pre@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    ***

Patrick J. Loll writes:
 > 	I'm trying to measure a weak anomalous signal at a relatively long
 > wavelength.  I have a high symmetry spacegroup (#96), and was able to take
 > LOTS of data--a continuous phi scan of 1440 deg (4 complete 360 degree
 > rotations).  Scaling all of the data together (or even selected subsets,
 > such as data from one 360 deg rotation) gives high R values, not surprising
 > in light of the expected significant absorption effects (data from the same
 > crystal taken at 12 keV scale beautifully with low R's).
 > 	Because I have so much data, I can scale in P1 to obtain a data set
 > with reasonable redundancy with decent agreement statistics.  I'd like to
 > take the scaled, unmerged data in P1 and apply a sliding box type of local
 > scaling, averaging all the Bijvoet pairs down to the asymmetric unit in the
 > process.   I think this should give the best treatment of the absorption
 > effects.  I can't use the Matthews & Czerwinski protocol, since their
 > derivation assumes a comparison of only two quantities (eg, F+ and F- or
 > Fnati and Fderi).  This yielded a nice simple non-iterative algorithm for
 > calculation of the scale factors, but is not appropriate for this case,
 > since I would like to average over all 16 of the symmetry-related
 > reflections.
 > 
 > Questions:
 > 
 > 1.  Does this sound nuts?  It seems to me that if one chooses a
 > sufficiently large box, the simultaneous estimation of even a large number
 > of local scale factors might remain robust.
 > 
 > 2.  Can anyone recommend a program to do this? Or toss some snippets of
 > code my way?
 > 
 > Thanks for any bright ideas.
 > 

Scala in the CCP4 suite has a number of useful options to do this. The
latest version (ftp://ftp.mrc-lmb.cam.ac.uk/pub/pre/scala_2.7.1.tar.gz
or in the imminent CCP4 release 4.0) has a spherical harmonic scale
parameterisation which should work well in this case. It will probably
only work with data integrated with Mosflm, since eg scalepack
declines to reveal the essential geometrical information in its
output.

	Good luck
	Phil