On Friday 21 September 2007 09:03:56 am Sue Roberts wrote:
> Yes, and I just got a pdb file back from the pdb that they flagged
> with over forty bad residues.
>
> The Molprobity plot from the RCSB validation server shows zero
> residues in disallowed regions. ( I got lucky, I didn't "fix" any bad
> residues.)
>
> For many of these "bad" residues, the position of the carbonyl oxygen
> is very clear in the electron density.  There are four copies of the
> protein in the asymmetric unit, NCS was never used in structure
> building or refinement, and all four copies of the residue have very
> similar psi-phi values - the conformations are real and there is
> nothing bad about the Ramachandran plot.
>
> Has anyone successfully fought this "bad residue" listing?  I
> understand that the pdb wishes to flag problem regions, but this is
> not a valid way of doing so.  I can tell them where the problem
> regions are in the structure if they wish.

The problem here is probably one of interpretation of the available
data.  Residues flagged as Ramachandran outliers are not necessarily
wrong.  They are simply Ramachandran outliers.

We as crystallographers should help our non-structural colleagues
to recognize that outliers do happen and the ones that are correct
are likely to be in interesting regions of the protein where the
protein scaffold is forcing regions into conformations that they wouldn't
otherwise want to adopt.  From my own work, uracil DNA glycosylase
has a conserved active site phenylalanine that's also a conserved
Ramachandran outlier.  I, for one, think that's pretty interesting and
have no doubt that I and others have determined that conformation
correctly.

The problem is that the PDB needs some automatic way to identify
"unusual features" as warning to the users of our structures.  These
do not indicate problems, but rather potential problems.

I had a similar conversation years ago about potential curation of
the metal binding sites in the Scripps Metalloprotein
Database.  While some people were advocates of manual curation,
the problem is that this simply doesn't scale.  But without
manual curation, things that are "different" can't be interpreted
as being "wrong".  They simply are "different".  But again, users
looking at metal coordination chemistry might like to know that
a Cu-S bond is two-sigma longer than normal.  It might be an interesting
feature of the protein, but it also might be a failure of the 
crystallographer to turn off VDW interactions during refinement
or a case where an inappropriate bond length was enforced
during refinement.  But I do believe that the users of the structures
ought to know if something is "different" so that they are aware
that they should be thinking about it.

Especially in your case, where there are multiple molecules in
the asymmetric unit with similar Ramachandran outliers, I can't
envision that this curation will cause anyone to think that
you got it "wrong".  We should welcome the PDB's
adoption of a more accurate definition of the allowed regions
of Ramachandran space.  It will cause all of us to look more
closely at interesting regions of our structures.

C.

-- 
Christopher Putnam, Ph.D.
Assistant Investigator
Ludwig Institute For Cancer Research

Reply via email to