Thank you very much for your help.

 

I just wanted to follow up in how I'm progressing with this.  I tried the things you suggested, as well as a variety of others, with no luck.  It turns out that my 4mol/AU P41 twinned data is ACTUALLY 8mol/AU P212121.  The packing in the AU has an almost 41 screw relating 4 mols to the other 4.  When this operation is carried out, 2 mols overlay 2 others almost perfect, while the other 2 overlay the other 2 by 180 degrees.  So my AU packing is a pseudo-41 with apparent twinning.  (The b dimension did not need halving, so my cell is 2x the size, if that makes sense).

 

The data refines very well as P212121, and Rfree drops like a rock.

Again, thank you so much for your suggestions, and Phenix is really wonderful!

Christina



----- Original Message ----
From: Pavel Afonine <[email protected]>
To: PHENIX user mailing list <[email protected]>
Sent: Tuesday, May 20, 2008 1:51:42 AM
Subject: Re: [phenixbb] problem with high Rfree

Hi Christina,

I usually observe similar R/Rfree behavior when I use LS target instead of maximum-likelihood. In fact, twin refinement in phenix.refine is based on LS target function. This is just to share my observations; it's not necessarily happening in your case.

- How many of test reflections you have and how they are distributed in resolution bins?
- Just out of curiosity: did you try to ignore twinning and run refinement with default target?
- What if you don't run SA?
- Did you try to play with target weights (try different values for wxc_scale and wxu_scale) to see how this affect the R-factors?
- Are you using the latest (or one of) phenix.refine?
- Did you try to fill in water?

Could you send a log file: may be we could spot something outstanding...

Pavel.


On 5/19/2008 1:32 PM, Christina Bourne wrote:
Hello BB,
I am working on a structure that is twinned (merohedral) in P41.  Initial estimates of twinning fraction (via xtriage and Yeates server) are around 0.37; in refinement they start around 0.49 and drop to 0.46 (so far).  My MR soln is good (Phaser LLG ~550) with 4 mols per AU; this gives good packing with no room for another mol.  I have data from 48-2.8A.
I should mention that data processing wasn't easy- this crystal form also has an apparent epitaxial twin in the b dimension causing a doubling in that dimension.  During indexing (with SAINT) I applied a transformation of 100 00.50 001 to the reflections, which then gave me the expected cell and from Xprep the data looks OK.  (Yes, I am trying to grow different forms)
Maps look nice, although I wish my ligand had more/better density.
 
I ran phenix.refine with defaults, and included twin_law="h,-k,-l", xray_data.r_free_flags.generate=true, detwin.map_types.aniso_correct=true and also simulated_annealing=true.  My starting R/ Rfree is 0.448/ 0.4385; ending is 0.2962/ 0.4270.  Geometry is somewhat tighter after refinement but is better (bonds 0.017 to 0.01, angles 1.966 to 1.530).  I have not included my ligand, and I can see a few spots where the model needs adjusting, but nothing majorly wrong.
 
Why is my Rfree hanging so high?  Is it becaues I have bad karma?  Is it to do with the twinning?  I am pretty sure my space group is right (based on the above info and the maps look fine; Phaser could not find a solution in P43).  There isn't room for anything else in the cell.
 
Any opinions or suggestions are most welcome as I wait for better xtals to grow...
Thanks in advance,
Christina


_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb