[phenixbb] Global metrics for Ramachandran quality beyond %outliers?

Dale Tronrud detBB at daletronrud.com
Thu Feb 5 11:13:24 PST 2015

Hash: SHA1

   The phi/psi plot is a particularly poor source of restraints
because it presents a global view, not a specific one.  The "outlier"
idea comes from a look at the blank spaces in a plot compiled from
many, many protein models.  Correctly configured residues in those
regions are rare (although they do exist).

   The problem with looking in the mass of the plot is exemplified by
the strong peak at the alpha-helix region.  In the average of many
models there are quite a few alpha helical residues.  There are,
however, proteins that contain no alpha-helicies.  Should a refinement
program try to turn some of the residues in such a protein into
alpha-helices just to match the global expectations?  Of course not.
The problem is that knowing the phi/psi angles of a residue in the
"allowed" region does not give you any clue as to whether that residue
would be better off a little to the right or a little to the left.

   If you noticed that a residue was in an alpha-helix but didn't have
the consistent phi/psi angles and you could nudge it closer without
hurting the fit to density much, that might be a good idea.  Context
is the key.

   There is a paper which points the way for doing this.
Hollingsworth, et al, (2012, JMB 416, 78-93) looks at successive pairs
of phi/psi angles and shows that they tend to cluster in 4D space.  It
might be useful to identify the motif of each residue and try to move
it toward the center of that motif's distribution.  Then you would be
making an alpha-helix residue a better alpha-helix and a type I turn a
better type one turn.  You model would have to be good enough to get
the motif right at the start and that limits the radius of convergence.

   A 2D phi/psi plot has too little information to make individual
restraints beyond simple "outlier" boundaries.  4D or 6D plots have
much more hope of success.

Dale Tronrud

On 2/5/2015 10:46 AM, Shane Caldwell wrote:
> I'm making a new thread because it's off the topic, but I have a 
> question in response to this point 
> <http://phenix-online.org/pipermail/phenixbb/2015-February/021691.html>:
>> [Doggy structure improved, as shown by how] The Ramachandran
>> plot
> tightened
> This is interesting to me - other than eyeballing the plot, are
> there any quantitative metrics of Ramachandran distribution
> quality? At low-res or when refining a bunch of structures
> side-by-side it could be an additional indicator of quality of the
> model, without requiring having to inspect the plot. Of course
> manual inspection is always important in the end, but in early
> stages a global measure of spread could help guide refinement.
> I've seen %outliers used for this purpose, but that ignores that
> while a residue can be "allowed" it can still be far from a
> statistically likely conformation. From what I've seen, some users
> only consult Ramachandran to tweak residues until they pop into the
> "allowed" regions and stop there, which isn't the same as globally
> improving the geometry.
> Interested to hear thoughts (or it it's already in use, pointing me
> in the right direction!),
> Shane Caldwell McGill University
> On Thu, Feb 5, 2015 at 5:06 AM, Andreas Förster
> <docandreas at gmail.com <mailto:docandreas at gmail.com>> wrote:
> Dear Almudena,
> I promise not to make a habit of advertising other programs on the 
> Phenix mailing list, but just once I would like to encourage you 
> (and the community) to try all the tools at your disposal.
> Different refinement programs have different strength and 
> weaknesses. Secondary-structure restraints in Phenix are great for 
> low-resolution data, but so is jelly-body refinement in Refmac.  A 
> dodgy 3.2 Å structure I'm currently working on was improved 
> dramatically by Buster-TNT.  The Ramachandran plot tightened, and
> R free plunged by four percentage points.
> Andreas
> On 05/02/2015 10:44, Almudena Ponce Salvatierra wrote:
> Dear all,
> I am refining my structure (data at 3 A), with a model that is 
> complete. However the Rs values are: R work= 0.25 and Rfree= 0.32.
> I have read "Improved target weight optimization in phenix.refine"
> (In the computational crystallographic newsletter 2011) and what I 
> understand is that just by marking the boxes "improve
> xray/stereochemistry weight" and "improve xray/adp weight" it
> should work... giving me the best possible Rfree.
> I'm refining individual coordinates, occupancies, b-factors 
> (isotropic for all atoms), TLS, and using secondary structure
> restraints, automatic ligand linking and experimental phases
> restraints. Also, I chose this strategy because I have finished
> building the structure and according to some of the suggestions in
> "towards automated crystallographic structure refinement with
> phenix.refine".
> I am actually quite confused and don't know what to think... is it
> a matter of the weights? is it only that this is as good as it
> gets?
> Any suggestions and comments are welcome.
> Thanks a lot in advance,
> Best,
> Almudena -- Almudena Ponce-Salvatierra Macromolecular
> crystallography and Nucleic acid chemistry Max Planck Institute for
> Biophysical Chemistry Am Fassberg 11 37077 Göttingen Germany
> _________________________________________________ phenixbb mailing
> list phenixbb at phenix-online.org
> <mailto:phenixbb at phenix-online.org> 
> http://phenix-online.org/__mailman/listinfo/phenixbb 
> <http://phenix-online.org/mailman/listinfo/phenixbb>
> _________________________________________________ phenixbb mailing
> list phenixbb at phenix-online.org
> <mailto:phenixbb at phenix-online.org> 
> http://phenix-online.org/__mailman/listinfo/phenixbb 
> <http://phenix-online.org/mailman/listinfo/phenixbb>
> _______________________________________________ phenixbb mailing
> list phenixbb at phenix-online.org 
> http://phenix-online.org/mailman/listinfo/phenixbb
Version: GnuPG v2.0.22 (MingW32)


More information about the phenixbb mailing list