[phenixbb] Ramachandran and clashes

[Nathaniel Echols]

On Tue, Jul 23, 2013 at 8:35 AM, Felix Frolow <mbfrolow at post.tau.ac.il> wrote:
> Were Ramachandran restraints in PHENIX reside now in the recent GUI's ? Once
> there were immediately visible as well as rotamer restraints.

Yes, I took them off the main interface because we (Jeff and I) found
that a) people were using them far too aggressively in circumstances
where they really shouldn't, and b) they tended to screw up local
structure and almost always result in chemically insensible
Ramachandran plots.  I really don't think there's any excuse to use
except in absolute desperation, and I actually think it would be
better to deposit a structure with 1% outliers than to use the
restraints.  Usually if the overall weight on the geometry restraints
is sufficiently high, the Ramachandran statistics will naturally get
better.

You can still find the parameter either by clicking the "Model
interpretation" button, or by using the parameter search (in the
Settings menu).  But I would exhaust every other possible option
before resorting to this.

We never had rotamer restraints (although I guess the new RSR method
uses them internally), just the fix_rotamers option, but that is now
merged with the RSR code.

> According to your mail from yesterday I wonder how can I treat Ramachandran
> outliers in very large molecules (close to 9000 residues).

Turn on weight optimization - this will almost always reduce both the
number of Ramachandran outliers and the clashscore.  The alternative
is to set wxc_scale much lower, but I've always found that manual
adjustment of the weights ends up taking far more manual effort and
just about the same amount of time as the automatic optimization.

-Nat