On Wednesday 18 April 2007 07:40, Phil Jeffrey wrote:
> Harry M. Greenblatt wrote:
>
> > You should be refining an overall temperature factor at that
> > resolution. It's one of the choices in the list, instead of "isotropic".
>
> I disagree with this. At that (3.2 Angstrom) resolution I've often
> found than a tightly restrained individual B-factor refinement gives a
> significantly lower R-free than a single overall B-factor. I also
> prefer it to grouped B-factors in CNS
My suggestion (as you can probably guess):
Step 1
Set the geometric restraints very tight, but the thermal restraints
not so tight. Refine individual or group B's (I doubt it matters which).
Step 2
Take that model, unrealistic as it is, and feed it to the TLSMD server
for automated analysis of the distribution of B factors in real space.
Tell the server to use the backbone B's only (it's a tick-box on the
'advanced options' page). The output should suggest to you which portions
of your structure are suitable for treatment as coherent groups;
some will be relatively well-ordered, others may be waving around wildly.
Step 3
Go back to refmac and perform a multi-group TLS refinement using the
groups suggested by TLSMD. Because this is low resolution, you should
refine a pure TLS model. That is, reset all the individual B's to
some constant value and let the TLS model alone describe their variation.
Success is never guaranteed, but in test cases we have found that for
low resolution structures a multi-group pure TLS model can give better
R values than individual isotropic B refinement, while ensuring that
the distribution of B's adheres to a physically plausible model.
Ethan Merritt
> , because the latter are not
> geometrically restrained and show a lot of physically unreasonable
> waywardness (although often, similar R-free as B-individual).
> Individual B's can also be restrained by non-crystallographic symmetry
> and as far as I can tell grouped B's are not.
>
> I think one has to explore all possibilities rather than take one fixed
> approach to working at modest resolutions, and the optimal solution is
> likely to be different for different structures.
>
> Phil Jeffrey
> Princeton, NJ
>
>
>
> >> Hi,
> >>
> >> I have a little problem with B-factor refinement. I'm using the CCP4i
> >> interface, Refmac 5.2.0019, a resolution of 30-3.2 A (I tried 8-3.2 A
> >> as well, it doesn't make a big difference for this problem), and a
> >> current Rfree of 30.4%.
> >>
> >> Refmac refines the B-factors so that they are nearly the same for main
> >> chain and side chain, and I don't like that (or could it make sense in
> >> any way?). Moreover, my structure is a protein complex, and Refmac is
> >> mainly doing this for one component of the complex. If I take the
> >> B-factors from the original uncomplexed protein (around 18, 1.75 A)
> >> and add 44 to them with moleman to get them in the range they are in
> >> the complex, Refmac "flattens" them remarkably in only 5 cycles of
> >> restricted refinement. Does anyone have an explanation for this? I am
> >> pretty sure that the complex components are in the right place, I see
> >> beautiful density and everything I should see at this resolution.
> >>
> >> Here is what I tried further:
> >>
> >> * I de-selected "Refine isotropic temperature factors" in the Refmac
> >> interface. There was no REFI BREF ISOT any more in the com file. But
> >> there was also no difference in the B-factors compared to when there
> >> _was_ REFI BREF ISOT in the com file... So does Refmac just _ignore_
> >> my wish not to refine B-factors? (The REFI keywords were as follows:
> >> type REST - resi MLKF - meth CGMAT - is there any B-factor-thing
> >> hidden in this?)
> >>
> >> * I played around with the geometric parameters. If I select the
> >> B-factor values there (the keywords are TEMP|BFAC
> >> <wbskal><sigb1><sigb2><sigb3><sigb4>), it does not make _any_
> >> difference, what values I fill in there, the resulting B-factors are
> >> always the same (but different from when I don't use the TEMP keyword,
> >> and even "flatter"). Default for WBSCAL is 1.0, I tried 10, 1.0, 0.1,
> >> 0.01, and the equivalent numbers for the sigbs.
> >>
> >> Thanks for any thoughts on this,
> >>
> >> Eva
> >
>
--
Ethan A Merritt Courier Deliveries: 1959 NE Pacific
Dept of Biochemistry
Health Sciences Building
University of Washington - Seattle WA 98195-7742
|