[phenixbb] phenix real space refine & free R reflections

[Tristan Croll]

This is essentially what I do by default when making real-space maps to 
work with. The relevant portion(s) of my maps.params file for 
phenix.maps would typically look like:

    map {
      map_type = 2mFo-DFc
      format = xplor *ccp4
      file_name = None
      fill_missing_f_obs = True
      grid_resolution_factor = 1/4.
      region = *selection cell
      atom_selection = None
      atom_selection_buffer = 10
      sharpening = False
      sharpening_b_factor = None
      exclude_free_r_reflections = True
      isotropize = True
    }

Cheers,

Tristan

On 2017-05-05 15:54, Edward A. Berry wrote:
> On 05/05/2017 05:01 AM, Hillen, Hauke wrote:
>> Dear Phenix bb,
>> 
>> I have a rather conceptual question regarding the treatment of free R 
>> flags in phenix.refine and phenix.real_space_refine. I am refining an 
>> xray structure using phenix.refine (but not using the real space 
>> refinement implemented there) and at some point decided to play a bit 
>> with the standalone program phenix.real_space_refine. In particular, I 
>> was interested in testing real space refinement against different maps 
>> (experimental map vs. model-phased). Being used to the excellent 
>> handling of free R flags in Phenix, I assumed giving 
>> phenix.real_space_refine the output MTZ of a previous phenix.refine 
>> run as a map would be safe to do. But I then saw the phenix.refine FAQ 
>> states the following:
>> 
>> "Are R-free flagged reflections included in the maps used for 
>> real-space refinement?
>> No, this is almost guaranteed to bias R-free; these reflections are 
>> removed internally prior to map calculation. However, the output maps 
>> will include these reflections unless you explicitly request 
>> otherwise.”
>> 
>> While it makes sense to me that including the free reflections in the 
>> maps used for real space refinement will bias R-free, I wonder why the 
>> default behavior of phenix.refine is to output maps which include 
>> these reflections. I assume most users will use these output maps (for 
>> example via the OPEN IN COOT button in the GUI) to iteratively improve 
>> their model by local real space refinement in Coot. Conceptually, how 
>> is real space refinement performed by phenix.refine (where the free R 
>> reflections are not used to calculate the map to refine against) 
>> different from manual local real space refinement in Coot? Will the 
>> latter not also bias R-free when the default output maps from 
>> phenix.refine are used?
>> 
> 
> I would argue just the opposite- free reflections should be included
> in the maps used for RS refinement. Excluding free reflections is the
> same as setting them to zero amplitude, so you are still biasing these
> reflections, just biasing them toward a drastically wrong value (zero)
> instead of biasing them toward fobs.
> 
> If the evidence for bias is that R-free gets "better" when you include
> the free reflections, isn't it more because you avoid biasing them
> toward zero than because you do bias them toward the Fobs? In the case
> of density modification, which has the same problem, Ivan Rayment*
> showed the importance of using the "fill-in" method for missing
> reflections in making maps. That would seem to be the correct way of
> implementing R-free for density modification or real-space refinement:
> you don't give away your Fobs, but at the same time you don't keep
> forcing the free reflections toward zero.
> *Acta Crystallogr B. 1983 Aug;39(4):505-516.
>  Acta Cryst. (1983). A39, 102-116
> 
> eab
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb
> Unsubscribe: phenixbb-leave at phenix-online.org