Re: [ccp4bb]: R-sym I/sigI etc...

["Eleanor J. Dodson" <ccp4@ysbl.york.ac.uk>]

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

Bart Hazes wrote:
> 
> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> Yes the best stratety is to include all information however with final shell
> statistics of R-sym 60.7 and I/sigI 0.9 it is reasonable for a reviewer to
> question if there really is "information" in the final shell. If this is a
> structure determination paper you can use the agreement between Fcalc and Fobs
> at high resolution as an argument to include this very weak data. For a
> crystallisation paper you don't have this luxury and may be better of to quote
> that diffraction with average I/SigI better than 2 was observed to ???
> Angstrom resolution even though you should use higher resolution data for
> refinement. You really just want to indicate the quality of the diffraction
> data.
> 
> What you should do with the paper is up to you to decide.
> 
> Bart
> 
> On Thu, 1 Aug 2002 rams@poori.biochem.uiowa.edu wrote:
> 
> > ***  For details on how to be removed from this list visit the  ***
> > ***          CCP4 home page http://www.ccp4.ac.uk         ***
> >
> >
> > Hello All,
> >
> > I thought this was a settled issue.  That R-syms and mean I/sigI's are a
> > non -ussue and that the best stratety is to include all information.
> >
> > I sent a crystallization manuscript with
> > R-sym in final shell 60.7 Redundancy 8.1 I/sigI 0.9 9all in the final
> > shell).
> >
> > Acta D refused to publish it (a crystallization manuscript) unless I cut
> > my resolution or my redundancy  to get better R-sym at the outermost shell
> > or cut I/sig I at 1.5..
> >
> > The catch is as you collect data to this redundancy the data does get
> > weaker and I/sigI gets poorer. I personally think this fine as we have
> > solved the structure refined etc.. etc.. and everything looks fine.
> >


 There is very good evidence that with Maximum Likelihood targets even
VERY weak data ( Rsym > 60%)  helps improve the model, providing it has
not got systematic errors. If I were the referee I would prob ask for a
Wilson plot to check whether the data seemed statistically valid, and
once it is solved, plots of R and FreeR against resolution..

 Various Murshudov et al papers on ML  demonstrate that the data can be
useful..

Eleanor