[phenixbb] close contact and torsion angle

Nathaniel Echols nechols at lbl.gov
Fri Oct 12 04:58:39 PDT 2012

On Thu, Oct 11, 2012 at 8:09 PM,  <hchxia at ujs.edu.cn> wrote:
> There are some close contacts (hydrogen bonds) in my refined model. Can I just ignore it, delete it or what?  I added hydrogen for refinement since the resolution is 1.47 A.  Should I remove all these added hydrogens and refine again?
> Second question is there are several outilers of torsion angles (6 from Ramachandran plot in coot ). But 16 are outside expected Ramachandran regions from the validation report from PDB bank(ADIT) . What is the cause?  And should I re-refine to correct them? Or just go ahead to deposit it?

As Pavel indicates, it's difficult to answer this without more
context.  But here are several partial answers:

1) Close contacts (in the way the term is used by Probe/Molprobity)
are usually okay, hydrogen bonds are good.  Significant overlaps (>
0.4Å) are bad (when viewing Probe output in Coot or KiNG, these will
be the jagged pink lines).  But it's very difficult to remove these
entirely; really severe problems need to be fixed, but if the overall
clashscore is less than 10 (for this resolution I would aim for < 5)
the model is basically okay.  If Pavel's suggestion doesn't work, and
your clashscore is very low, I wouldn't worry too much - but I can't
really say for sure without seeing the specific clashes.

1b) Assuming you're at the end of refinement anyway, leave the
hydrogens on (they will improve your R-factors at this resolution) try
setting nonbonded_weight=200 and re-refining with XYZ weight

2) Phenix is using very new Ramachandran distributions from the
Richardson lab based on a larger set of reference structures, which
changed the boundary between "allowed" and "outlier" slightly.  Right
now it may still disagree with the Molprobity web server (hopefully
this will change very soon); I have no idea what the PDB uses.
Additionally, slight differences in the way values are stored and/or
interpolated can determine whether a residue is flagged - this is why
Phenix often disagrees with Coot on corner cases.  Regardless, I would
trust the output of Phenix over most other sources, as it is based on
the best statistics the Richardson lab has been able to obtain so far,
sampled very finely.

As always, if you are able to show us the model and data (OFF-LIST -
just email Pavel and/or me), we may have a more detailed answer.


More information about the phenixbb mailing list