Re: [ccp4bb]: B factors

[Edward Berry <eaberry@lbl.gov>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***

Randy Read wrote:
...
> Actually, in this equation mu (usually "u") refers to the component of the 
> displacement in the direction parallel to the diffraction vector (i.e. 
> perpendicular to the corresponding Bragg planes).  If we assume isotropic 
> thermal motion, then there will be similar displacement in three orthogonal 
> directions, so to get the radial rms displacement in any direction we have to 
> multiply by the square root of 3.  The result is that a radial rms 
> displacement of 1A comes at an isotropic B-factor of about 26A^2, not 79.
>
Thanks for the clarification, Randy. That's what I get
for being too lazy to go through the derivation.
The revised table:

B=26 ->     1 A* rms deviation <U^2>
B=79 ->     1.73 A*
B=100 ->    1.94 A*
B=104 ->    2 A*.
B=234 ->    3 A*
B=416 ->    4 A*
B=650 ->    5 A*

Even with the correction, because of the sqrt dependence on B
the B factors around 100 do not lead to huge rms displacements.

Vondelft, Frank wrote:
> disorder) than having to believe that one atom's positional uncertainty
> relative to its neighbours is in the order of its bondlengths to those
> self-same neighbours(*), which makes for somewhat awkward chemistry.
> Our residual Bfactors do indeed end up in that range, even when the
> WilsonB is 50+ (and I don't think that's simply the result of the
> default in ccp4i).
> 
> (*) Yes, indeed: isotropic INDIVIDUAL Bfactors...

I would have thought the value we refine as individual B-factor
says nothing about its motion being correlated or not with that
of its neighbors, and of course if the rms movement is around 2A
then it must be correlated with its neighbors, or bonds would
be breaking?


wgscott@chemistry.ucsc.edu wrote:
 > I have a 3 Angstrom resolution structure that refined in CNS with a
 > <B>=78 A**2 and an Rfree of almost 30% and using Refmac <B>=42 A**2 and
 > Rfree of 23% (keeping the same test/working set division and starting
 > model and similar geometric deviations from ideality).
 >
 > I worry that something might be going wrong with the CNS bulk solvent
 > model in CNS, or that there might be some other problems.

Over my head again, but I seem to remember some discussion that the
Babinet solvent model used by refmac led to a strong correlation
between solvent B and model B so that for low resolution structures
one or the other had to be fixed, and the value of the other depended
on what you chose for the one? I thought the CNS/Xplor solvent model
gave more reliable B values at low resolution.

I may be very confused about all this,
in which case I would appreciate some more
clarification.

Ed