[phenixbb] Difference Map positive peaks in release dev-798

Pavel Afonine pafonine at lbl.gov
Mon Jun 27 10:43:31 PDT 2011


Hi Yuri,

> I was wondering if anything changed (weights or something) in the way 
> the difference maps are calculated?
> I just ran refinement on a 1.3 structure and a  hexacoordinated Mg2+ 
> ion has significant positive density peak even at occupancy 1.0. I had 
> not noticed that before.
> My ligand also does have positive peaks, but refine gives it a .91 
> occupancy.
> I wonder if I am missing some parameter  cif file or some definition 
> that would make the positive peaks decrease?

thanks for sending the data!
I did refinement the way that I believe is the most appropriate given 
the data and model quality:

- the command I used:
phenix.refine params --overwrite

- the parameter file contains these lines:

refinement {
   input {
     pdb {
       file_name = model.pdb
     }
     xray_data {
       file_name = data.mtz
       r_free_flags {
         file_name = data.mtz
       }
     }
     monomers {
       file_name = ligand.cif
     }
   }
   output {
     prefix = test
   }
   refine {
     adp {
       individual {
         isotropic = (element H or element D)
         anisotropic = not (element H or element D)
       }
     }
   }
   main {
     ordered_solvent = True
     number_of_macro_cycles = 10
   }
   target_weights {
     optimize_wxc = True
     optimize_wxu = True
   }
   ordered_solvent {
     n_cycles = 2
     new_solvent = isotropic *anisotropic
   }
}

Since you sent me a .sca file that did not contain the R-free flags (and 
NOT the actual MTZ file that you used to refine your structure), so I 
had to do one of the worst possible things: re-create Rfree flags from 
scratch (note: Rfree flags should be created once, before any refinement 
or model building is done, and never changed since that). To make the 
refinement results more or less meaningful, I had to do the usual things 
to remove bias from Rfree: reset ADPs to isotropic, remove water, shake 
coordinates and do as many refinement macro-cycles as necessary to 
achieve convergence (10 in this case, see above). Also, I add H atoms. 
This is what model.pdb is.

After refinement I got good R-factors (given the resolution): r_work = 
0.1158 r_free = 0.1268 (the starting model you sent me has: r_work = 
0.1430 r_free = 0.1595. And now I'm looking at your MGs:

- Fig 1:
http://cci.lbl.gov/~afonine/tmp/tmp_09357452/p1.png
My interpretation of this is that MG probably partially occupies two 
sites, so I would try replacing that water (that phenix.refine 
automatically placed) with another MG, and refine occupancies of both MG 
(independently or constraining them to the sum = 1). Alternatively, 
could it be that you simply have two partially occupied waters here?

- Fig 2:
http://cci.lbl.gov/~afonine/tmp/tmp_09357452/p2.png
In this case one of the waters needs to be shifted towards green 
density, and you need to add another water to empty blob (I don't know 
why phenix.refine did not add it automatically - I will investigate once 
I get a chance). A slight negative peak at around MG center may indicate 
under-refinement of its ADP, the undefined charge (you probably need to 
define it in input PDB file), may be its partial occupancy, etc... 
Eliminating it would probably require a lot of experimenting with 
different possibilities.

I will send you the PDB and MTZ files in a separate email (OFF-LIST).

Pavel.



More information about the phenixbb mailing list