[phenixbb] How to reduce clashscore value

Nathaniel Echols nechols at lbl.gov
Fri Nov 18 06:58:06 PST 2011

2011/11/17 CelinaSocek <oceanlinerth at hotmail.com>:
> I'm now using phenix to refine a small protein about 300 residues long. The
> resolution of my data is about 2.9 angstrom. Currently I got a refined model
> with good R-factors about 23/26, RMSD very reasonable and no Ramanchandran
> outliers. However,  I found that the clashscore is extremely high: 63!! And
> rotamer outliers are also a lot: 8.5%. I wonder, what kind of stretagy I can
> use to reduce this? Thank you very much!

Which version are you using?  We made some changes over the summer
that improve the geometry, especially the clashscore - these are
present in version 1.7.2.  If you're still using an earlier version,
this might largely fix the problem.  However, if there is
conformational strain in the model, it is difficult to fix severe
clashes without further rebuilding.  You should inspect the clashes in
Coot (the Phenix GUI has shortcuts for this, but Coot alone also has
this capability) and fix as many as possible yourself.  It is
recommended that you use explicit hydrogen atoms when doing this,
since the reason for clashes is often not obvious if only heavy atoms
are used.  (You don't necessarily need to refine with heavy atoms
though - they're just an aid to visualizing clashes.)

The rotamer problem may be related; there isn't much you can do with
these except fix them manually, but relieving some of the clashes may
help.  The dihedral angle restraints simply aren't that strong, so in
areas of weak density it can be difficult to completely prevent


More information about the phenixbb mailing list