[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

weak reflections

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

Hi everybody,

I'm really wondering why this topic comes up periodically
in the mailing list. It has been discussed many times
already, and there seems to be only one sensible solution
and that is to NOT exclude any weak reflection from the
refinement. It will make the structure worse (!) but it will
make the R-factor look better. If you don't believe this,
read Hirshfeld & Rabinovich, Acta Cryst A (early 80s).

If a sigma cutoff is applied to the data, more weak
reflections will be omitted at higher resolution than
at low resolution. That means that the average F changes
differently dependent on resolution, that means that
the slope in the Wilson plot is changed and that leads
to a systematic error in the thermal parameters. How
that affects the positional parameters, I don't know,
but it does.

Concerning the resolution limit, the parameter of choice
is certainly I/sigma. The merging R-factor is inversely
related to I/sigma (in fact, Rmerge = 0.7979 * sigma/I when
only statistics is taken into account and the redundancy is
really high), plus it's dependent on the redundancy
of the data. A good criterion for the high resolution limit
seems to be I/sigma > 2.0 and at least 50% of all reflections
observed in that resolution bin. A high merging Rfactor
therefore does not necessarily mean that the data is noise.
One has to be very careful with that statement anyway.

Happy Wednesday,



 Manfred S. Weiss, Dr.
 Institute of Molecular Biotechnology
 Department of Structural Biology and Crystallography
 Beutenbergstrasse 11                                 Fon: +49-3641-65-6081
 Postfach 100813                                      Fax: +49-3641-65-6062
 D-07708 Jena                                    Email: msweiss@imb-jena.de
 GERMANY                        Web: http://www.imb-jena.de/www_sbx/manfred