It is in general matter for debate what to do in these extreme cases.
When NCS rotation is parallel to the potential twin operator even if
twin is present electron density does not change much (at least not as
much as we would like). There are many cases where because of this
twinning was not noticed at all. However it did not change electron
density or conclusion you drawb from the derived model.
To compare twin and non-twin you need to do sufficient number of
cycles of refinement and model building. Then your R/Rfree may
indicate differences better.
When you do twin refinement you add only one parameter (in case of
hemihedral twinning: in general the number of extra parameters is
equal to the number of twin domains minus one). I am not sure
overfitting is a danger here.
Twin fraction 43% indicates that there might be strong twinning (when
twin fraction is refined to a value close to 50% then it could be
because of higher symmetry or twinning)
To refine properly in the cases like yours you need to select
reflections carefully accounting for rotational symmetry (twin or
otherwise). For example you can generate free reflections in higher
space group (e.g. P41212) and then expand to lower symmetry. This way
you make sure that twin (or ncs rotation) related reflections belong
to the same class (working or free). If you have not done this then
freeR may be misleading. refmac (and I am sure that other programs
also) try to make sure that twin related reflections belong to the
same class. But if you do not specify twin then two ncs rotation
related reflections may belong to two different classes and it may
obscure Rfree. In your case it seems that only in the P41212 case free
and working reflections are not related with symmetry.
Until you have done all these tests (making sure that free is generate
in higher space group, sufficient number of cycles of refinement and
perhaps model building) it would be difficult to say if you have
twinning. Looking at the statistics you give it seems that there is
not enough evidence to reject hypothesis that space group is P41212
( I do not consider freeR since they do not seem to be "free" enough)
I hope it helps
regards
Garib
On 23 Jun 2009, at 02:16, Brett Collins wrote:
Dear all,
I posted here recently regarding a potential twinning problem I had
and got some very useful feedback.
The summary of what happened is I had a crystal that appeared to be
P42212 from systematic absences after scaling in SCALA (a,b,c;
118,118,192). Interestingly POINTLESS suggested P41212 (or P43212 as
the best solutions), and PHASER could only find a solution in P41212
(2 mol/asu). Cumulative intensity plots suggested twinning, as did
phenix.xtriage, and refinement at 2.6 A resolution was stuck at R/
Rfree of 25.5/32.0. (Rmerge 6.5%). Although people suggested using
refmac with twin refinement, in P41212 no twin laws were identified
in REFMAC or phenix.xtriage.
I tried indexing in lower symmetry P222 (a,b,c; 118,118,192) (Rmerge
5.4 %). PHASER found a solution in p212121 (4 mol/asu). Refinement
now proceeds to R/Rfree of 25.3/27.9.
I tried additional twin refinement in in P212121 using REFMAC, which
found the twin operator k,h,-l with fraction ~43%. However, R/Rfree
only changed marginally (and suspiciously) to 25.6/26.9.
Importantly, in each case the maps seem good and I can't discern any
significant difference between them.
My questions are
(i) which (if any) solution should I trust?
(ii) Is the drop in R factors in p212121 just because parameters/
data ratio is less? i.e Am I just overfitting the data?
(iii) Do I really have twinning?
Cheers,
Brett
################################
Brett Collins, PhD
Group leader
Institute for Molecular Bioscience
Level 3 North
Queensland Bioscience Precinct
The University of Queensland
St. Lucia, 4072, Qld,
AUSTRALIA
e-mail: [email protected]
phone: 61-7-3346-2043
FAX: 61-7-3346-2101
website: http://www.imb.uq.edu.au/index.html?page=82433
Courier address:
Institute for Molecular Bioscience
Queensland Bioscience Precinct
Building 80, Services Road
University of Queensland
St. Lucia, Brisbane
Queensland, Australia 4072
################################
