[phenixbb] discrepancy in MolProbity statistics

Nathaniel Echols nechols at lbl.gov
Mon May 14 18:49:29 PDT 2012

On Mon, May 14, 2012 at 6:37 PM, Michael Thompson <miket at chem.ucla.edu> wrote:
> I recently ran into a discrepancy between the MolProbity statistics reported by phenix.refine and the MolProbity statistics reported by the actual MolProbity webserver. I was hoping someone could comment on this and let me know if this is a warning sign that something is amiss. The structure (resolution=2.0A) seems to be correct and well refined, based on R-factors (0.1591/0.1774) and geometry reports from the various validation tools.

Did you refine with explicit hydrogens, and if so, did you skip the
step in the MolProbity server where you add hydrogens?  The likely
explanation is this:

1) You ran phenix.ready_set (or phenix.reduce) to add hydrogens, which
were placed at the nuclear distances (approximately 1.1A).
2) phenix.refine shortened the C-H bonds to conform to X-ray
scattering distances, i.e. the center of the electron cloud,
approximately 1.0A.
3) when you run validation in Phenix, the hydrogens are stripped off
and re-added internally, again at 1.1A, and this temporary model is
used to calculate the clashscore.
4) when you take the final model with explicit hydrogens and submit it
to MolProbity, the hydrogens will all be at X-ray distances, which
means they will be as much as 0.2A further apart from each other than
they otherwise would be.  Since the clashscore counts any overlap >
0.4A, this has a significant effect.

So, as a general rule, when running Molprobity you should always let
it add new hydrogens.  (I'd recommend using the Phenix validation
tools anyway, as they will be more consistent with the refinement
output, although I suspect both are doing something dumb with
hydrogens on ligands.)  The number reported by Phenix is the one you
want to report in your publication, and since the statistics are
spectacular for this resolution it won't be a problem.  (I'm not sure
about the rotamer discrepancy.)

(And yes, someday we'll fix this - the Richardson lab is working on
determining appropriate hydrogen parameters that can be used
consistently for both refinement and validation.  But since Phenix
also does neutron refinement, this is another one of those problems
that is much more complicated than it initially appears.)


More information about the phenixbb mailing list