[phenixbb] Twin Law question

Pavel Afonine pafonine at lbl.gov
Thu Oct 28 12:55:31 PDT 2010

  Hi Young-Jin,

> As my recent data did not give me perfect statistics, I was wondering what the problem was (one of things is ice ring though).  When I ran Xtriage with my struggling current data set and looked like I got twinned data.  I reran PHASER as was suggested and confirmed that current space group is right.  My question is now how I can handle this problem.  First, as was recommended I input twin_law, however not all the cases (I have a couple of data sets) were working. One favorite(!) error message was as following:
> "Map type 'Fc' not supported for twinned structures. Allowed types: Fo-Fc, Fobs-Fobs-Fmodel, 2mFo-DFc, 2mFobs-DFmodel, mFo-DFc, mFobs-DFmodel, gradient, m_gradient."
> I got this from both GUI and shell environment.

in the command line all you need to to give phenix.refine the twin law:

phenix.refine model.pdb data.mtz twin_law="-h,l,k "

as it is explained in the manual:

If you are using the GUI then also you can input the twin law  
information, and if it then crashes then Nat is the best person to 
explain it.

> 1. Now from now on, whenever I want to phenix.refine, should I include twin_law command such as 'twin_law=-h,l,k'?


> 2. In this case, what do I expect to see such as lowering Rfree or else? If it doesn't help lowering Rfree,

Typically, after discovering twinning and taking it into account the 
R-factors should drop by 3-10% or so.

> what to do next?

I guess just finalize your model and make sure it's right -:)

> 3. Do I have to report I did with twinlaw option in case of preparing a paper or presentation?

Yes. And the relevant information is printed in REMARK 3 records of your 
refined model. IMPORTANT: make sure this information is preserved when 
you PDB deposit your model.

> How much can I convince my refinement is OK with twined data?

The rock-solid way is to show result of refinement with and without 
considering twining (showing R-factors may be enough).

I almost forgot... Once you decided to use twinning make sure the Free-R 
flags are generated in PHENIX so you will not get biased Rfree.

> Additionally, I checked other data, and applied twin_law but the result R/Rfree were exactly same as the one I refined without twin_law option, it this normal case?

If using twin information doesn't change anything then it's not worth of 
using at all.

Finally, run:

phenix.model_vs_data mode.pdb data.mtz

Does it tell your data is affected by twining? If not - forget about it.


More information about the phenixbb mailing list