[phenixbb] fix Ramachandran outliers

Nathaniel Echols nechols at lbl.gov
Tue Jul 30 10:49:05 PDT 2013

On Tue, Jul 30, 2013 at 10:28 AM, Pavel Afonine <pafonine at lbl.gov> wrote:
> This is because Coot and Phenix use slightly different Ramachandran plot
> libraries. I guess Paul Emsley or/and Jeff can offer more comments on this.

Both are actually derived from MolProbity, but there are two differences:

1) The phi/psi distributions in Phenix 1.8.1 or newer come from an
expanded set of high-resolution structures ("Top8000" instead of
"Top500"), and also have separate distributions for Ile/Val and cis
versus trans-Proline.  (This should be consistent with the MolProbity

2) The distributions in Coot are stored in 10-degree increments,
versus 2-degree increments in Phenix.  Interpolation is used to
estimate the actual frequency for exact phi/psi values, but this is
less accurate with a coarser distribution.  So I would trust the
numbers in Phenix over the ones in Coot.  (I have no idea what the PDB
is using at this point, but I suspect you'll get a completely
different set of numbers, as usual.)


More information about the phenixbb mailing list