[phenixbb] Molprobity percentile statistic
Nathaniel Echols
NEchols at lbl.gov
Tue Oct 20 11:12:00 PDT 2009
On Oct 19, 2009, at 7:03 PM, hari jayaram wrote:
> And thats the reason I liked the molprobity statistics expressed as
> percentile since it helped me realize how far I have to go relative to
> other structures at the PDB. If its not too difficult can I cast my
> vote to report it inside the phenix GUI. Its just that running the
> full analysis inside of phenix is significantly faster than running it
> on the duke webservers.
It isn't very difficult at all - but we need to discuss the best
source of these scores. An additional complication in Phenix is that
we also incorporate diffraction data into validation (depending on
which version of the program you're using), which complements the
geometry analyses. One side effect is that only PDB entries for which
structure factors are available are included in the distributions used
by Polygon. (This probably improves the distributions, however.)
> Regardless thanks for the hint about phenix.polygon .
> I just looked at my phenix.polygon report and I am afraid its a little
> asymmetric.
Since the server bounced the original post I've saved the image you
sent here (since it's a good example):
http://cci.lbl.gov/~nat/img/phenix/polygon_example.png
The colors are a bit confusing (next release will be improved), but
this is about what I'd expect to see. What you should look out for is
values that fall on the extremed ends of the distribution, or very
lopsided or pointy shapes. I can't really suggest absolute guidelines
since these statistics are so different and the PDB won't (and
shouldn't) have a Gaussian distribution for each one - and keep in
mind that most models in the PDB would probably benefit from further
refinement. I find this type of analysis very informative, but we
don't want people to fall into the trap of leaving obvious errors in
the model because they only looked at global scores.
Anyway, I've added some more similar graphs for surveying PDB
statistics in the GUI - they'll be in the next version.
-Nat
-------------------
Nathaniel Echols
Lawrence Berkeley Lab
510-486-5136
NEchols at lbl.gov
More information about the phenixbb
mailing list