Hi Dirk,

Nice cross validation flags for twin refinement do obey the twin law:
if (h,k,l) is in the free set, then  (k, h, -l) should be as well.
Same is/should be true for work set btw.

In phenix, we generate test reflections that follow the lattice
symmetry (*). This means that even if you do not realise that your
data is potentially twinned (or has a wrong point group), you do not
end up biasing your test set. The actual reductionof a low symmetry
free set to high symmetry free set is a bit of 'art- and fly-work'
(kunst en vliegwerk) , but is doable.

The CNS "make_cv_twin.inp" script needs a specific twin law. In the
case where more then a single twin law is possible, taking them all
into account is not a bad idea.

Cheers

Peter



(*) check out:

http://phenix-online.org/cctbx_sources/cctbx/cctbx/miller.py

and look for the function

def denerate_r_free_flags_on_lattice_symmetry




2007/10/23, Dirk Kostrewa <[EMAIL PROTECTED]>:
> Hi John,
>
> if I understand the "make_cv_twin.inp" correctly, then this script takes
> care about separating reflections related by the twin law to avoid a
> potential bias of the Free-R set with twin-law-related reflections from the
> working set. At least, in the source of this script I read the comment: "{-
> make sure all twin related reflections are in the same set -}". If this is
> true, then you simply don't have to worry any further about your good
> R-factors, but enjoy a very well refined structure!
> Maybe, the authors of this CNS script could comment on this separation of
> reflections?
>
> Best regards,
>
> Dirk.
>
>
> Am 23.10.2007 um 15:32 schrieb john kryst:
>
> Hi CCP4bb ,
>
>              Thanks for your suggestions... here is the summary of the
> structure...
>
> I am working in deletion and point mutants.  I have a mutant (P61 resoln
> 3.1A) with partial twinning with twinning fraction 0.28. Wilson b-factor is
> 50. The data collected in the home source with normal exposure time. The
> rmsd's of the refined structure  is 0.0081 (bonds) and 1.38672 (angles).
> Molprobity is score is 50 without any manual model building (just after the
> refinement).
>
> Since the structure is twinned new free R-flag has been chosen according to
> CNS make_cv_twin.inp. The b-factor has been flattened to 30. Then
> rigid-body_twin refinement followed by simulated annealing_twin is done
> (heating the structure up to 5000K with different initial velocity trials).
> The R and R free after this step is  18.35 and  27.22. Then i did
> bgroup_twin refinement which gave me  R and R free 17.06 25.63. Further
> model building and refinement tend to reduce the R and Rfee upto 15 and 23.
>
> I deleted some portion of the molecule and did refinement. But proper
> density appeared  after it.
>
> Please help me out.
>
> John
>
>
>
> *******************************************************
> Dirk Kostrewa
> Gene Center, A 5.07
> Ludwig-Maximilians-University
> Feodor-Lynen-Str. 25
> 81377 Munich
> Germany
> Phone:  +49-89-2180-76845
> Fax:  +49-89-2180-76999
> E-mail: [EMAIL PROTECTED]
> *******************************************************
>
>

Reply via email to