[phenixbb] Phenix.refine distorts geometry during rigid body refinement

[Nathaniel Echols]

On Wed, Sep 14, 2011 at 11:35 AM, Ian Slaymaker <iamslaymaker at gmail.com> wrote:
> I am bringing a PDB refined in Refmac back to phenix.refine. After an
> initial rigid body refinement to compare numbers (one chain per
> group), the geometry (specifically the bond length rmsd) had
> surprisingly increased. Phenix.validate confirmed the Refmac (initial)
> geometry numbers and then afterwards confirmed the distorted geometry.
> What is strange is that when opened in COOT, the initial and final
> models are completely super imposed.
>
> REMARK Start: r_work = 0.3375 r_free = 0.3624 bonds = 0.006 angles = 0.768
> REMARK Final: r_work = 0.3379 r_free = 0.3626 bonds = 0.145 angles = 1.019
>
> can you help me figure out what is going on here? These are the final
> few refinements of a 4.1 A structure.

The problem is with your rigid body group definitions: only residues
1-7 are included, which means they will move independently of the rest
of the chains.  Since the rigid-body refinement ignores geometry
restraints, it will pull these residues as far apart as necessary,
leaving you with bond deviations up to 16A.  We probably need to check
for this during refinement and alert the user; even if you are running
restrained refinement afterwards, huge deviations from ideal restraint
values can screw up the automatic weighting.

All that aside, I wouldn't run rigid-body refinement here anyway - if
Refmac and Phenix disagree on the optimal positions of the atoms, it
will be much more subtle than domain-level movements, which is what
the rigid-body refinement is for.

-Nat