[phenixbb] questions about twinned structure and MR

Peter Zwart PHZwart at lbl.gov
Fri Apr 18 11:43:07 PDT 2008

Hi Christina,

> I am new to Phenix and am facing my first twinned structure.  While I wait
> for my copy of Yeates Methods in Enzymology (albeit from 1997) to be
> delivered, perhaps this bb could answer a few questions?

There are a number of excellent papers besides the one you mention.
Check for instance
Dauter, Z. (2003). Acta Cryst. D59, 2004-2016.
Parsons, S. (2003). Acta Cryst. D59, 1995-2003.
Zwart,Lebedev, Mushudov,Adams   Acta Cryst. (2008). D64, 99-107

> At what point do I detwin, or do I not do it at all?  It seems Phenix does
> this itself, on the fly?  Is it refining the twin fraction too?

No need to detwin really. Maybe helpfull for density modification, but
typically it does more harm then good. phenix.refine does refin ethe
twin fraction as well indeed. Please provide data as obtained from
your data reduction program without detwinning it yourself.

> MR (Phaser) finds solutions with twinned or detwinned data.  However, the
> solutions vary by 180degrees, so I suspect I'm finding solutions for both
> h,k,l and h,-k,-l.

Quite possible.

> These come from two different runs, but one has refined
> LLgain of 1010.8 (matches soln found with detwinned data, both run through
> ccp4), and other has refined LLgain of 1073 (through Phenix; it's rotated
> 180degrees from the other).

How many copies do you have in the ASU?

> When performing AutoBuild after AutoMR, should I worry that I haven't
> assigned Rfree?  Does Phenix account for twinning when it does this
> automatically, or what keyword(s) do I need to include?

It should assign free flags automatically in a sensible manner when
using the autobuild wizard. Please provide the wizard with a set of
parameters so that it knows that your data is twinned. It will use
that in refinement, but not in density modification.

> I see a section in the documentation about refining twinned structures, and
> am looking forward to getting to that point!  I just want to be sure I'm
> following a "correct" procedure for the steps leading up to that.  From the
> literature, it seems quite common to not account for twinning until
> refinement.  My concern is the two MR solns...which should I use?  If I know
> it's twinned, shouldn't I account for that from the beginning?

Just pick the best one. If you pick the 'twin related solution' your
twin fraction could refine to a value larger than 0.5.

> More details:
> P41, data good to 2.2A, 2 molecules in the AU, twinning fraction ~0.4.
> Rmerge from scaling is reasonable at ~11% overall.  MR model is 100%
> identical to this data, but from a different spacegroup (thus I'm running
> MR).
> Thanks for any tips or guidance!
> Christina Bourne

The dimer in the ASU is probably the reason why you pick up two
possible solutions.
The dimer NCS axis is most likely parallel to the twin axis (two fold
along a). So in this case, the twin related solution and NCS related
solution might be the same.

The twin tutorial in the phenix documentation should provide some useful info.



More information about the phenixbb mailing list