[phenixbb] question on bad geometry

[Pavel Afonine]

  Hi,

I agree with Tom. Here is an example.

A 3ifk structure deposited in PDB (resolution 2.03A) has the following 
statistics:

- reported Rwork/Rfree = 0.237/0.293
- recomputed Rwork/Rfree (using phenix.model_vs_data) = 0.2306/0.2835
- Molprobity scores (computed using phenix.model_vs_data):

Ramachandran plot, number of:
   outliers : 4     (2.35  %)
   allowed  : 1     (0.59  %)
   favored  : 165   (97.06 %)
Rotamer outliers        : 28 (18.67 %)
Cbeta deviations >0.25A : 10
All-atom clashscore     : 61.61 (steric overlaps >0.4A per 1000 atoms)

After running this model through PHENIX AutoBuild and then refining it 
with phenix.refine using fix_rotamers option and H atoms added, I'm 
getting the following numbers:

- Rwork/Rfree = 0.2097/0.2402
- Molprobity scores:

Ramachandran plot, number of:
   outliers : 2     (1.18  %)
   allowed  : 2     (1.18  %)
   favored  : 166   (97.65 %)
Rotamer outliers        : 5 (3.57 %)
Cbeta deviations >0.25A : 1
All-atom clashscore     : 37.14 (steric overlaps >0.4A per 1000 atoms)

As you can see, the change is pretty significant and no manual work at 
all to achieve it!

Good luck!
Pavel.

P.S.: All files are here:
http://cci.lbl.gov/~afonine/example_3ifk/

3ifk.mtz - original data file
pdb3ifk.ent - model from PDB
pdb3ifk.mvd - result of "phenix.model_vs_data 3ifk.mtz pdb3ifk.ent > 
pdb3ifk.mvd
" command
autobuild.mvd - result of "phenix.model_vs_data autobuild.pdb 3ifk.mtz > 
autobuild.mvd
" command
autobuild.pdb - model after AutoBuild and phenix.refine


On 8/31/10 7:08 PM, Thomas C. Terwilliger wrote:
> Hi Fengyun,
>
> That does look like a lot of outliers to me.  You could try improving the
> model by "rebuilding in place" with autobuild with the keyword
> "rebuild_in_place=True" which will try to rebuild your model without
> changing the sequence alignment. This procedure takes a while and it is
> useful to use a multiprocessor machine with nproc=5.
>
> You should also have a careful look at your model, as some manual
> rebuilding might improve it a lot.
>
> All the best,
> Tom T
>
>>> Hi everyone,
>>>
>>> I have a structure at 2.3 A refined in phenix to R/Rfree=0.23/0.28.
>>> However, the geometry is not that good, the followings are the output
>>> from phenix.model_vs_data,
>>>
>>>       Stereochemistry statistics (mean, max, count):
>>>         bonds            :   0.0080   0.0547 3244
>>>         angles           :   1.1669  12.1085 4374
>>>         dihedrals        :  15.7241  85.6280 1202
>>>         chirality        :   0.0706   0.2831 536
>>>         planarity        :   0.0036   0.0246 573
>>>         non-bonded (min) :   2.1217
>>>         Ramachandran plot, number of:
>>>           outliers : 17    (4.02  %)
>>>           allowed  : 37    (8.75  %)
>>>           favored  : 369   (87.23 %)
>>>         Rotamer outliers        : 30 (8.62 %) goal:<  1%
>>>         Cbeta deviations>0.25A : 0
>>>         All-atom clashscore     : 40.51 (steric overlaps>0.4A per 1000
>>> atoms)
>>>
>>> I tried to use fix_rotamer=True, but it doesn't help at all. As to the
>>> rotamer outliers, the value is far higher than the ideal one. Does
>>> that mean I need to manually adjust the model? Or will autobuild help
>>> improve the geometry?
>>>
>>> Thanks!
>>> Fengyun
>>>
>>> _______________________________________________
>>> phenixbb mailing list
>>> phenixbb at phenix-online.org
>>> http://phenix-online.org/mailman/listinfo/phenixbb
>>>
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb