Re: aniso scaling

[Edward Berry <eaberry@lbl.gov>]

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

S.Ravichandran wrote:
>         Well, if you look into the text books on crystallography, the
> simple equation for R-factor calculation involves a 'scale factor' (k)
> which is 'always' applied to Fcalc.
> 

Maybe we can get the discussion on B-factors going again. I don't 
know if it matters to which F the scale factor is applied, provided 
you're not going to refine occupancy. But I want to argue that, 
since you probably are going to refine atomic B-factors, the B-factor 
should always be applied to the Fcalc in  scaling, and the 
isotropic part should be passed on to the atomic B's in the
coordinates file. I'd be interested  to hear what others think.

Let's say you collect a lousy dataset that
has average B of 60. You make your map, build your model, and
the program probably assigns initial values of 20 for all B-factors.
Now you start refinement, scale your Fobs against the Fcalc which
will have B-ave of 20, and suddenly your dataset has an overall 
isotropic B-factor of 20. When you submit the coordinates they 
will still have average B-factor of 20, but this gives a 
completely false impresion of how well ordered the crystal was.
The B-factor increases F in the hi-res shells, but increses sigma(F)
by the same factor.

OK, there may be valid reasons for adjusting the B-factor of the
F-obs, for making maps or Patterson or rotation function searches, 
sharpening the data to accentuate the hi-res details, but the final 
atomic B's should be refined against an uncorrected dataset.

Having said this, I realize that all the current software handle
this correctly, and what I've said is obvious to experienced
crystallographers, so maybe there is no need for discussion. But
in the Xplor manual, including the online version, there is
an example file which intentionally and with malice aforethought
swaps Fobs and Fcalc before scaling so that in effect Fobs gets
scaled:

>Example: Refinement of Overall Anisotropic B-Factors
>
>Prior to invoking the refinement and after the refinement, the Fcalc 
>and Fobs arrays are swapped. This implies that the B-factor
>correction is applied to the Fobs. The corrected Fobs are then written 
>to a new reflection file. To refine an isotropic B-factor, simply set
>the ANISotropic option to FALSE, and write the coordinates with the 
>modified B-factors to a file instead of writing the scaled Fobs. 
>
<a href="http://xplor.csb.yale.edu/manual/xtalrefine/baoverall.inp";> 
baoverall.inp </a>

OK, now I see this is done only for the anisotropic component, which 
cannot be accomodated by isotropic atomic B's. But if one applied this
routine when the average atomic B is very unreasonable, as with a new 
model with arbitrary B, that B-overall apparently would get transferred 
to the Fobs, and I think there should be some warning.  

I suggest a new REMARK for the PDB format:

REMARK 787: Atomic individual B-factors were refined against Fobs which
REMARK 787:  had not been adjusted by any applied B-factor and therefore
REMARK 787:  reflect the true state of order in the crystal.

	or

REMARK 787: Atomic B-factors were refined against a dataset which had been
REMARK 787: scaled with application of an arbitrary isotropic B-factor.
REMARK 787: Therefore B-factors are an indicator of relative order of
REMARK 787: different parts of the model, but the average B is meaningless.


Ed

---
PS- I am not a crystallographer by training, and I have said some pretty
silly things on this bulletin board before. If this is another such case,
... . . I'm sorry