[phenixbb] Real Space Refinement with marginal data

Pavel Afonine PAfonine at lbl.gov
Wed Apr 14 17:08:39 PDT 2010

Hi Carsten,

> I was wondering if anyone had experience/advice for real space 
> refinement with phenix using marginal data to 3A. I just did a run 
> (SA, real-space refinement against diff_map, 2 fold NCS, grouped APD, 
> fix_rotamers enabled) were the resulting structure showed lots of 
> disconnected sidechains and some broken main chains. I am just 
> repeating the run w/o real space refinement, may be the combination of 
> SA and real space refinement was a bit harsh,  but am wondering what 
> kind of parameters make most sense to tweak in order to avoid having 
> the molecules fall apart.

this could be a weight issue (weight between real-space target and 
restraints). In general I didn't find it too beneficial doing real-space 
refinement globally, although I've seen cases where it made a good 
difference (but at higher than 3A resolution). You may try to switch to 
"simple" target (to make it run much faster) and turn on the weight 

  real_space_refinement {
    mode = *simple diff_map
    target_weights = value *grid_search gradients_ratio
    rmsd_max_bonds = 0.05
    rmsd_max_angles = 5
    grid_search_scales = 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.5 3

"fix_rotamers=true" (local real-space refinement for the residues that 
really need it, as determined automatically) is much better option. I 
thoroughly tested it up to 3A resolution and in all test cases it worked 
just great. Try it and see what happens for your data.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20100414/eadbc7f9/attachment-0003.htm>

More information about the phenixbb mailing list