unrefineable structure ?

[Zbyszek Otwinowski]

schiweck at argon.biophys.mpg.de wrote:

> processing with Laue symmetry P41212 gives a Rmerge (about 6.9 %) which
> is only slightly higher than the one I get for Laue symmetry P41 (6.5 %).

98% chance that it is P41212, 2% chance that it is a merohedral twin,
with 2 molecules in a.u.

> Rigid-body refinement in P43212 for the whole molecule or the
> individuals domains results in a drop from 48 to 42 % for the R-factor
> and from 47 to 44 %
> in Rfree. In the following refinement the R-factor drops to about 30 % whereas
> the Rfree first drop to about 41 %

One should stop refinement here and evaluate result

> and then after the R-factor has reached a
> value of about 35 % rises again to a final value of 45 % ! This behaviour is
> independent of the refinement protocol.

It means that all refinement protocols are wrong in this stage, as
refinement
goes away from correct solution (it is antirefinement)

> Even with torsion angle dynamics
> which should solve my problem of underdetermination, (99.3 % completeness to
> 3 A but still about 12000 parameters to refine with 9000 reflections for 10 to
> 3 A) I cannot refine my structure.

Do you have 12000 torsional angles? 12000 seems to refer to something
else.

> As everybody can imagine I checked the molecular replacement solution again
> and still feel that it is correct.

It seems so.

> Does anybody know about a similiar case where low-resolution data
> (which of course are not brilliant but reasonable) were not sufficient for
> refinement

Merohedral twinning will allways produce bad refinement statistics, in
such
cases either non-twinned crystal is found or project is abandoned.
Merohedral
twinning does happen, but it is rarely reported.

> or where this kind of overfitting occured.

overfitting is very freequent, in significant fraction of papers,
particularly
in papers that do not report r-free.

> Any suggestions are welcomed.

Change significantly refinement protocol, check input for errors,
refine tigthly restrained b-factors as part of the molecule maybe
quite disordered and b-factors may correct for that. Sometimes
b-factor refinement (tigthly restrained!) can help a lot at 3.0A

  Zbyszek Otwinowski                     |        zbyszek at chop.swmed.edu
  University of Texas                    |        tel : (214)-648-5098 
  Southwestern Medical Center            |        fax : (214)-648-5095 
  5323 Harry Hines Boulevard  Dallas, Texas  75235-9038
[EOB]