[phenixbb] Refinement

Pavel Afonine pafonine at lbl.gov
Wed Mar 25 18:56:58 PDT 2015

Hi Mohamed,

> I have a structure at 1.4 A currently with an Rfactor of 21/26 %.

R-factors are a bit high... At this resolution one would expect Rwork 
~15-17 and Rfree ~16-21 (lookig at statistics for similar resolution 
structures in PDB).

Did you add water, refine anisotropic ADPs, add H, model alternative 

> I notice from the plot by cycle that everytime the RMS bonds/angles 
> reduce, the Rfactor increases a few percent. Is there any way of 
> telling phenix.refine to not 'restrain' the model too much?

Try running refinement with weights optimization - usually this 
addresses this issue.

> I have already used automatic weighting.

Oh I see you did.

> However, as I have an NCS of 4, this can be tedious to do by hand, and 
> AFAIK, phenix.refine does not create new conformations. Is there an 
> alternative?

I'm not aware of software that builds alternative conformations fast and 
reliably. Make sure you do not use NCS restraints - at this resolution 
using them may be counterproductive.

> Final question, some residues were flagged as having high B factors 
> but not in all the copies. How do I deal with this?

At this resolution it's expected to observe differences

> Some background: I initially had problems with refinement being stuck 
> at 30-35 % in SG P 3 2 1 (as suggested by Pointless and Xtriage). 
> However, the beamline scientist suggested that there could've been 
> lattice translocation and that it is actually P 3. Reintegrating seems 
> to have solved the problem. With integrated data up to 1.2 A, I tried 
> cut-off at 1.4, 1.3 and 1.2 A. Together with CC1/2 and refinement 
> results, I chose 1.4 A, but should I cut it a bit lower?

I see.. Crystal oddity may contribute to higher (than anticipated) 
R-factors. If you modeled everything you can model, model to map is ok, 
and model passes all other validation then perhaps there isn't much else 
you can do.

More data you cut lower R factors you may get but a) R-factors computed 
using different sets of reflections are not comparable, and b) is 
lowering R really ultimate goal?


More information about the phenixbb mailing list