Rfree & Bijvoet pairs
Gerard 'CD' Kleywegt
GERARD at XRAY.BMC.UU.SE
Thu Mar 21 02:58:14 EST 1996
The problem is more general than that
Due to bulk solvent, every reflection will be related to its
immediate neighbours (G-function), and in the case of non-
crystallographic symmetry, also to N other reflections
in the same resolution shell and their neighbours, related
by h' = R(T) h, where R(T) is the transpose of the
rotation matrix of an NCS operation
I've experimented quite a bit with different ways of
selecting test reflections, including small spheres
of test reflections (no NCS), and thin resolution
shells of test reflections (with NCS)
Using these different methods in two extreme cases
showed very little difference in fact:
- an intentionally backwards-traced structure with
4-fold NCS
- a published structure with a spacegroup error
In the first case, any method of chosing the test
reflections reveals the error; in the latter case
*none* of the methods reveals the error (note that
the relationships are exact in the case of missed
crystallographic symmetry); the actual numbers (i.e,
values of Rfree) vary only little and the difference
are probably within the error limits
Nevertheless, it *feels* safer to select the reflections
in spheres or thin shells ;-)
DATAMAN (soon to be released by the CCP4) contains
options to select test reflections in many different
ways
--gerard kleywegt
uppsala
More information about the X-plor
mailing list