[phenixbb] wxu scale in phenix.refine

Pavel Afonine pafonine at lbl.gov
Tue Sep 18 09:45:57 PDT 2007

Yes, I'm copying the response for exactly the same question that I wrote 
a week ago or so...

Part 1:

Playing with wxu_scale is how you change the strength of ADP restraints 
in all cases except when you do combined "TLS+individual ADP" 
refinement. In order to make the TLS refinement very robust and 
efficient we re-refined ~350 TLS containing structures from PDB 
performing the grid search for wxu_scale parameter and we found some 
"optimal" values for wxu_scale as a function of resolution. We use these 
values in automatic adjustment that you are mentioning. So currently 
there is no way you can turn off this adjustment; however I realize that 
users may want to play with wxu_scale themselves and skip the 
adjustments so I will provide this option in future.

Part 2:

As a possible solution to keep going now you can try this two-step 
step 1: Refine your model with default strategy 
(strategy=individual_sites+individual_adp) making sure that the ADP 
restraints are tight enough (play with wxu_scale) to produce the 
B-factors distribution that you like.
step 2: Refine the best model out of "step 1" with the 
strategy=individual_sites+group_adp+tls. This will preserve the 
distribution of B-factors in the model from "step 1" and will add TLS 
model on top of it.

Please let me know if you have any questions/problems!

Jan Abendroth wrote:
> Hi Pavel,
> thanks a lot for your fast response!
> Here some clarification:
> - somehow, the adjustments of the B weights were not taken into 
> account at all. Ie. a "diff" on the pdb-files from jobs run with 
> wxu_scale=0.5 and wxu_scale=2.0 shows that they are identical. While 
> the log file acknowledges the "Command line parameter definitions" for 
> wxu_scale=1 and wxc_scale=2, during the refinement section it states 
> that wxc_scale = 1.0 and wxu_scale=2.16.
> - yes, I included a TLS refinement. Are TLS refinement ans wxu_scale 
> mutually exclusive? How would one proceed in this situation?
> Cheers
> Jan
> On 9/18/07, *Pavel Afonine* <pafonine at lbl.gov 
> <mailto:pafonine at lbl.gov>> wrote:
>     Hi Jan,
>     > in the final stages of refining a structure, I would like to
>     optimize
>     > the restraint weights for geometry and B factors.
>     Good idea! Although phenix.refine computes "optimal" weight
>     automatically, the values vary from structure to structure.
>     > It seems to work fine for geometry by defining
>     > "refinement.target_weights.wxc_scale=1.0 ".
>     Yes, I would try an array of values 0.25, 0.5, 1.0, 1.5, 2.0, 2.5.
>     > However, when the respective definition for B-factors does not
>     seem to
>     > have an impact at all.
>     This is very weird. We exercise this in out tests regularly, so it
>     changing wxu_scale should have an impact. It may be that values range
>     for you structure is somehow different... Can you try the extreme
>     values: very big (wxu_scale=10) and very small (wxu_scale=0.1) and see
>     if this makes a difference? Are you using TLS (if yes, then this
>     behavior is explainable)?
>     > The main reason why I try to fiddle around with the B-weights is
>     that
>     > coot gives me a whole bunch of red bars in the Temperature factor
>     > variance analysis...
>     The variability of ADP depends on many factors: resolution,
>     model  used
>     for global motion (TLS), how internal degrees of freedom modeled ... I
>     do not know if Coot analysis takes this all into account. Otherwise I
>     would use the "red bars" analysis with some care.
>     Pavel.
>     _______________________________________________
>     phenixbb mailing list
>     phenixbb at phenix-online.org <mailto:phenixbb at phenix-online.org>
>     http://www.phenix-online.org/mailman/listinfo/phenixbb
> ------------------------------------------------------------------------
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://www.phenix-online.org/mailman/listinfo/phenixbb

More information about the phenixbb mailing list