Dear Kay,
your argumentation might be true if you compare refinements at
different resolutions. However, in your case, you refine a structure
at a given resolution. Any overfitting will affect _all_ data up to
this resolution. Looking at an R/R-Free at any lower resolution is
very different from refining a structure at that lower resolution
with its worse observations-to-parameter-ratio! Therefore, if I'm
right, any discrepancy in the R/R-free plot vs resolution at any
lower resolution point than the maximum resolution can't be simply
explained by arguing that at this resolution the amount of
overfitting would be higher. The actual differences between R and R-
Free against resolution must be due to other sources, like the usual
measurement errors, incomplete & inaccurate models, NCS influence,
and so on.
Best regards,
Dirk.
Am 15.11.2007 um 17:41 schrieb Kay Diederichs:
Dear crystallographers,
today I explained to a student that I believe that the difference
of R_work and R_free should decrease as a function of resolution,
because at high resolution there is less danger of overfitting,
whereas at low resolution one is always overfitting to some extent.
I'd say this should be true at least in absolute terms, but
probably even in relative terms: for example, for a 1.5A structure
I'd expect R_work/R_free= 14%/16%, whereas at 3A I'd expect 28%/35%.
I believe that I saw plots (maybe in a link given in a posting on
CCP4BB) obtained for structures from the PDB, which confirmed this
hypothesis. But I've been googling and searching through postings
now for quite some while, and cannot seem to find anything to that
effect.
Does anybody have a pointer to an analysis of this effect?
thanks,
Kay
P.S. Would be good to put this into the CCP4 Wiki article on R-factors
--
Kay Diederichs http://strucbio.biologie.uni-konstanz.de
email: [EMAIL PROTECTED] Tel +49 7531 88 4049 Fax 3183
Fachbereich Biologie, Universität Konstanz, Box M647, D-78457 Konstanz
*******************************************************
Dirk Kostrewa
Gene Center, A 5.07
Ludwig-Maximilians-University
Feodor-Lynen-Str. 25
81377 Munich
Germany
Phone: +49-89-2180-76845
Fax: +49-89-2180-76999
E-mail: [EMAIL PROTECTED]
*******************************************************
