[phenixbb] 0 for wxc_scale and wxu_scale?
pafonine at lbl.gov
Wed Nov 26 08:29:08 PST 2008
which version of PHENIX you are using? Please make sure you are using
the latest version.
> I am refining a structure with a resolution of 2.1 Angstroem using Phenix.
> The spacegroup is P 31 2 1 (No. 152) with a twin fraction of 0.37 and twin
> law -h,-k,l.
> In the refinement procedure the NCS (4 chains) constraints are on, using
> individual sites, ADP and occupancy refinement.
Just a suggestion to quickly try out... What happens if you turn off the
NCS restraints and do some refinement without using NCS (with and/or
without optimizing the weights)? I know it depends on many factors, but
I've seen a good number of cases at around 2A resolution where the
refinement was better if NCS is not used.
> In the last refinement step, I use the option to optimize the values for
> wxu_scale and wxc_scale. These values are optimizes to 0 for wxc_scale and
> close to 0 for wxu_scale (0.2)
It's all relative, the absolute values don't tell you much. If the R-,
R-free and Rfree-R are better after wxc_scale and/or wxu_scale
optimization and the geometry makes sense, then you got better model.
> Interestingly I obtain better R values (0.1948/0.2184), than without the
> optimization by using the default values (0.1391/0.2438 Rwork/Rfree).
Clearly, (0.1948/0.2184) is much better than (0.1391/0.2438). Some 10%
gap between R and Rfree at 2.1A resolution is bad.
> How is it possible that wxu_scale and wxc_scale can be optimized to 0 by
> having better statistics?
> The total refinement target is defined as
> Etotal= wxc_scale x wxc x Exray + wc x Egeom
> As I understand it, the crystallographic refinement target (Exray) would now
> not be considered in the refinement procedure.
> Would this refinement procedure be valid?
It doesn't happen in all macro-cycles and this is important.
You can also try to play with the weights manually by trying different
values for wxc_scale and wxu_scale.
> And how is wxc (output of target weights: x-ray data) defined?
See phenix.refine documentation page:
section "8. Target weights calculation". With some variations, the
implementation is very close to what's described there.
Please let us know if you have any other questions or problems!
More information about the phenixbb