Hi Ed,
DISCLAIMER: When I say "grouped B-factor refinement" I mean CNS-style,
Bmain/Bside refinement. Not to be confused with more general "domain
B-factor refinement" where single B-factor is assigned to some part of
the structure.
- In general you are free to decide what you name a domain: it can be a
residue, its part or the whole structure.
- What would be "main" and "side" for non-amino acid molecule, like a
whatever ligand?
That's exactly why ("effective number of parameters in presence of
restraints") individual B-factor refinement works well in phenix.refine
at lower resolutions, where people traditionally tempt to switch to
group isotropic ADP. Going to lower resolution, I only switch to group
ADP refinement if I really have to.
Just as it worked well in CNS and works well in REFMAC (which doesn't
even have the grouped B-factor refinement). It may be an extreme point
of view, but I think grouped B-factor refinement is useless distraction.
Unlike domain B-factor refinement coupled with thoughtful decision as to
which parts of the molecule may move together. Which again is not
necessarily better than refining one B-factor per atom with tight
restraints.
It boils down to similar point that tightly restrained refinement of
individual coordinates would be equivalent to rigid body refinement. In
practice, it is not. You can prove it by doing simple numeric
experiments. I did this sometime ago when working on this paper
(Automatic multiple-zone rigid-body refinement with a large convergence
radius. Afonine et al), but I didn't include the results since it was
out of scope of that paper.
Each data quality requires proper model parameterization. Otherwise, why
don't we do multipolar refinement at 2A resolution, or refine individual
anisotropic B-factors at 3A resolution? Simply because it would be
improper model parameterization given the data quality. It would be the
same nonsense as refining individual isotropic B-factors at 4A
resolution. Although technically, using proper weights, you probably
could -:)
I will add this
functionality to phenix.refine at some point.
It is rather obvious that I meant restraints between groups, not inside.
As for implementation of such restraints, it may be easier said than
done. While it's reasonable to apply the same restraints to every pair
of covalently bonded atoms, restraints between groups are not
necessarily identical.
Of course -:) That's why it's still not implemented in phenix.refine.
For instance, main-chain/side-chain restraints
in the same residue may need to be weaker than those applied to backbone
atoms when restraining neighboring residues. Side-chains on the surface
apparently will be less correlated with the backbone than those in the
core.
We cover this using "sphere-B-factor restraints". See (note, the formula
there has a few typos; see code code for the correct definition):
Afonine PV, Grosse-Kunstleve RW, Adams PD:
The Phenix refinement framework CCP4 newsletter No. 42, Summer 2005.
Potentially, the same concept can be applied to group B refinement.
I have a feeling that I've
seen/heard somewhere the idea of imposing B-factor restraints based on
interatomic distances, may be applicable here.
Yes, this is what is implemented in phenix.refine. See formula in the
reference above -:)
All the best!
Pavel.
