Re: [ccp4bb]: refmac fom

["Gerard \"DVD\" Kleywegt" <gerard@xray.bmc.uu.se>]

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


come on, guys - this is an ancient (non-)issue. back in 1996, two
"Rfree-cionados" observed:

"It is desirable to include all diffraction data when calculating
electron-density maps to avoid truncation errors. In principle, the use of
real-space refinement techniques could introduce some bias towards the test
reflections, but the seriousness of this effect has not been demonstrated.
Moreover, if simulated annealing is used throughout the refinement, any model
bias is likely to be removed during subsequent refinement." [*]

i'm still not aware of any convincing demonstration (and not even of an
unconvincing one, for that matter) of said effect, neither when it comes to
manual rebuilding or real-space refinement. i suspect it's a red herring (and
a sophistic one at that) :-)

so, rather than flaming him, i'm afraid i'm going to have to agree with bert
all the way on this one - sorry, guys ;-)

--dvd

[*] 
http://www.ncbi.nlm.nih.gov:80/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=8805582&dopt=Abstract


> > Also (Brent as well raised the question), qtfm (q for quoting):
> >
> > "Rebuilding into these 2mFo-DFcalc and mFo-DFcalc maps
> > seems to be easier than using classic nFO-(n-1)FC and difference maps,
> > consistent with the established technique for SigmaA style maps.
> > One advantage here is that since the m and D values are based on the Free
> > set
> > of reflections they are less biased than the values obtained by the CCP4
> > version of SIGMAA after refinement"
> >
> > Ok I can see that for the first (no previous rebuild) map - but then
> > in further cycles, haven't you actually taken info from your
> > crossvalidation set and (real space) refined against it?
> 
> Eleanor already indicated that the refmac map used D*Fc terms for Rfree
> reflections. This will make the purists happy, but I wonder if it is the best
> thing to do. Really, how big is the risk of introducing model bias by a clumsy
> human trying to best fit a piece of model into the density. We are not talking
> about clever software adjusting thousands of parameters to "force" a fit
> between model and observations. Perhaps lavishly adding waters to any positive
> density feature would introduce some bias but I hope us humans aren't THAT
> clumsy. The disadvantage of using D*Fc is that it doesn't contain ANY
> information about how to modify the current model to resemble the real
> structure more closely. It just contributes a weighted down echo of the Fcalc
> density. This may make the density a bit more beautiful and give the
> suggestion that your model fits the density better than it really does. So
> instead of model bias we create mind bias. Just imagine the silly example that
> you select 100% of reflections for the Free set leading to the use of D*Fc for
> all reflections. The map would probably look very good. My gut feeling is that
> the real purists should leave out the Rfree reflections completely and the
> practicalists would include them just like the working set reflection. The
> D*Fc option seems half-hearthed at best.
> 
> In Olympic tradition: Let the flames begin...

******************************************************************
                        Gerard J.  Kleywegt
    [Research Fellow of the Royal  Swedish Academy of Sciences]
Dept. of Cell & Molecular Biology  University of Uppsala
                Biomedical Centre  Box 596
                SE-751 24 Uppsala  SWEDEN

    http://xray.bmc.uu.se/gerard/  mailto:gerard@xray.bmc.uu.se
******************************************************************
   The opinions in this message are fictional.  Any similarity
   to actual opinions, living or dead, is purely coincidental.
******************************************************************