[phenixbb] bad geometry problem

[Pavel Afonine]

Hi Sarah,

thanks for sending me the data. I spent 15 minutes trying a few things 
and here is the statistics I get:

r_work = 0.2454 r_free = 0.2700
MOLPROBITY STATISTICS.
  ALL-ATOM CLASHSCORE : 13.60
  RAMACHANDRAN PLOT:
    OUTLIERS : 0.89  %
    ALLOWED  : 5.33  %
    FAVORED  : 93.78 %
  ROTAMER OUTLIERS : 8.64 %
  CBETA DEVIATIONS : 0

which isn't that bad compared to the starting point:

r_work = 0.2469 r_free = 0.2613
MOLPROBITY STATISTICS.
  ALL-ATOM CLASHSCORE : 48.87
  RAMACHANDRAN PLOT:
    OUTLIERS : 2.49  %
    ALLOWED  : 5.33  %
    FAVORED  : 92.18 %
  ROTAMER OUTLIERS : 29.47 %
  CBETA DEVIATIONS : 10

Here is what I did (it's just one of possible ways of doing this; other 
scenarios may works as good as this one or better!):

- add H atoms:

phenix.reduce model.pdb > model_H.pdb

- then run a quick geometry regularization:

phenix.pdbtools model_H.pdb --geom

- then strip off H:

phenix.reduce -trim model_H_regularized.pdb > start.pdb

- then run refinement using NCS and Ramachandran plot restraints:

phenix.refine start.pdb def_data.mtz main.ncs=true --overwrite 
output.prefix=def2 main.number_of_mac=5 ramachandran_restraints=true 
rama_potential=emsley strategy=individual_sites+individual_adp

- then take outcome of previous refinement and run the same refinement 
plus weights optimization:

phenix.refine def2_001.pdb def_data.mtz main.ncs=true --overwrite 
output.prefix=def3 main.number_of_mac=5 ramachandran_restraints=true 
rama_potential=emsley strategy=individual_sites+individual_adp 
optimize_xyz=true nproc=6

Also, looking at the original maps I see a lot of places where simple 
fixing in Coot is just obvious, for both - rotamers and Ramachandran 
outliers. So you should do it first of all.

Also, use Ramachandran restraints ONLY to preserve good starting model 
and NOT to fix problems.

All in all, geometry problems you were concerned about are all easily 
fixable. One of the reason why these problems arise is that the map is 
really looking like 3A resolution map, may be even worse; plus there are 
places in the model that you need to fix manually (refinement would not 
be able to fix them).

I will send you the files off-list.

Pavel

On 6/19/12 1:17 AM, Qian Sarah wrote:
> Hi, this may not be a new problem, but I tried every stretagy that I 
> know from bb, but the geometry is still very bad. So I wish I can get 
> some help from here.
> Crystal information:
>  SAD data. Resolution 3.0.  There is no anisotropy in the data.
>  R/Rfree : 22.7/26.8 (Rfree-flag is always from the Refine_data.mtz 
> file generated after the first refinement.)
>    RMS(bonds):0.02
>    RMS(angles): 1.453
>    Ramachandran outliers: 3.0%
>    Ramachandran favored: 91.2%
>    Rotamer outliers: 18.7% !!
>    Clashscore: 31.84
>    overall score: 3.47
>    I tried different strategies, like autobuild and then refine; set 
> wxu/wxc=0.3/0.3; turn on "Ramachandran restrains", "fix bad sidechain 
> romaters"etc, however the paramters are still very bad, especially the 
> rotamer outliers. Then I rebuild them manually in coot,after 
> refinement the parameters are even worse.
> p.s. phenix version: phenix-1.7.3-dev-1046.
> Thank you very much!
> best,
> Sarah