Hi Bjørn,
some things to consider:
- Usually when detwinning data some reflections are rejected (because they end up with negative intensities). This will make comparison of the results between the refinements difficult. It is worth looking at how complete your data are in the different refinements.
- The twinning estimate in xtriage usually is somewhat different from the refined twin fraction, which should in theory be more accurate (as it is refined using the data).
- As long as the intensity statistics look ok (i.e. the NZ plot and L-test look normal) for the detwinned data then it is reasonable to use the ML target. The ML target has assumptions that do not hold when the data is left twinned.
- You might want to judge which is better by looking at the quality of the density for a ligand or something else that you can leave out of the structure as a cross validation. I'd also look at the distribution of B-factors in the models for each case. In my experience these are what change the most when you account for twinning in the refinement.
Cheers,
Paul
On Sep 10, 2012, at 10:44 AM, Bjørn Panyella Pedersen
Dear list I would like some advice on how to do the final refinement of a twinned structure: I have a 3.0A structure (sg R3) which is twinned. I have collected ~50 datasets and the estimated twin-fraction varies from 0.2 to 0.45 based on phenix.xtriage analysis (britton plot etc). The structure could only be solved by experimental phasing using _detwinned_ datasets (detwinned by the ccp4 program DETWIN). Refinement of the model was initially done using my highest resolution _detwinned_ dataset (3.0A, britton_alpha:0.21) and a ML target in phenix.refine. Lately I have switched to th twin_lsq_f target and refined using the 'unmodified'/twinned dataset.
1. I have tried systematically to detwin the dataset with different twin-fractions and then refine using the ML-target and my best refinements comes using an alpha of 0.2 (see attached figure). This fits the Britton plot etc.
2. xtriage estimated the twin-fraction to by ~0.21, but using twin_lsq_f, phenix.refine now gets alpha to be 0.31. This seems like a large increase to me?
3. Using the manually detwinned dataset and a ML-target gives better R-factors than using the twin_lsq_f target (see attached figure).
Bottom line: How should I report/deposit this structure: A. Should I use the manually detwinned data and an ML-target to get a model that explains the data best (lowest R-factors). And deposit my structure with modified (i.e DETWINNED) structure factors. I fear that people would react strongly to the deposition of modified F_obs in the pdb. I could deposit both twinned and detwinned structure-factores but this might also confuse?
OR B. Should I use the twin_lsq_f target and accept that my deposited model is not as good as I know it could be. And deposit my structure with the real (i.e TWINNED) structure factors.
OR C. Something I'm not aware of means that refinement using the ML-target is not allowed on detwinned datasets, causing me to have artificially low R-factors.
I hope I'm not missing something basic here.:) Thank for any advice, pointers and ideas!
-Bjørn
-- Bjørn Panyella Pedersen, PhD
Macromolecular Structure Group Dept. of Biochemistry and Biophysics University of California, San Francisco MC2240, 600 - 16th Street S414 San Francisco, CA 94158-2517
Phone: +1 415-476-3937 E-mail: [email protected] http://www.msg.ucsf.edu
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
-- Paul Adams Deputy Division Director, Physical Biosciences Division, Lawrence Berkeley Lab Division Deputy for Biosciences, Advanced Light Source, Lawrence Berkeley Lab Adjunct Professor, Department of Bioengineering, U.C. Berkeley Vice President for Technology, the Joint BioEnergy Institute Laboratory Research Manager, ENIGMA Science Focus Area Building 64, Room 248 Tel: 1-510-486-4225, Fax: 1-510-486-5909 http://cci.lbl.gov/paul Lawrence Berkeley Laboratory 1 Cyclotron Road BLDG 64R0121 Berkeley, CA 94720, USA. Executive Assistant: Louise Benvenue [ [email protected] ][ 1-510-495-2506 ] --