[phenixbb] Refinement of perfect twined data of P31/P32

Nathaniel Echols nechols at lbl.gov
Mon Apr 22 07:57:52 PDT 2013

On Mon, Apr 22, 2013 at 7:45 AM, miyatake <miyatake at riken.jp> wrote:
> Now I'm triying to refine almost perfect twining data
> (mixing of P31 and P32). It has a twining operation of (h, k, l) = (-h, -k, l),
> detected by the xtriage program.
> The data qualigy seems to be good, deflacting to 1.6A resolution with Rmerge ~ 5%.
> When I refined the MR structures with phenix.refine, however, the model geometries have been worse as over the 20% residues fall into disallowed Ramachandran region,
> and Rfree stacked around 30%.
> I tried many options such as real space refinement, SA, bad rotamer fixation, NCS restrain,,.., but they seem to be not so effective as usual twininig cases..
> One molecule consists of 202 residues, and three molecules are inolved in AU.
> Of course, I used the twin option (-h, -k, l) during the refinement.
> Manual collection of the models could not improve the model anymore.
> Would anyone please advise me how should I refine the twinned data?

This sounds pathological (and/or a bug), but it's difficult to tell
without looking at the model and data - would you be willing to send
the files to us privately?  (I'm sure Pavel will want to see it too,
so help at phenix-online.org is the best address for this.)

One suggestion would be to run AutoBuild - with 1.6Å data it should
work very well (as long as you enter the appropriate twin law).


More information about the phenixbb mailing list