The topic brings up a question that I've been wondering about for some
time, perhaps someone can enlighten me. Why is it not standard practice
to deposit map coefficients along with structure factors ? Unlike image
deposition there are no significant storage or file format issues. This
would preserve a record of the "final" refinement used for publication,
bypassing the impossible task of recording/reconstructing the program
version and options used.
On 04/04/2014 09:13 AM, Gerard Bricogne wrote:
I replied in too great a rush - it is the Rfree, or the Rfree-Rwork gap,
that would go up a lot if a structure originally refined at low resolution
with targeting to an external high-resolution one was re-refined without
that targeting.
Gerard.
--
On Fri, Apr 04, 2014 at 04:55:03PM +0100, Gerard Bricogne wrote:
Dear Nat and Tim,
On Fri, Apr 04, 2014 at 08:18:06AM -0700, Nat Echols wrote:
On Fri, Apr 4, 2014 at 1:57 AM, Tim Gruene <[email protected]> wrote:
A more up-to-date reason is that programs calculate R values very
differently. If you take a PDB file refined with program X and put it
into program Y you easily get discrepancies greater than 5%.
This is actually pretty rare - usually it's only 1-2% at most.
Discrepancies like 16.5% versus 30.9% usually indicate that there's
something wrong or misleading in the annotation of the entry, and often
mean that you can't even reproduce the R-factor with the specified program.
-Nat
It could also be that such a structure is the result of a refinement
against low-ish resolution data that was restrained by "targeting" it to
another similar structure that was refined against higher-resolution data,
so as to retain the good local geometry of the latter in spite of the
shortage of data at low resolution. This is perfectly legitimate, and was
first done by Oliver Smart in BUSTER under the acronym of "LSSR", outlined
in an ACA 2008 Abstract then fully described in Acta Cryst. D68, 368-380
(2012). A similar feature has also become available in REFMAC and PHENIX.
The problem is that the deposition process is very likely to have lost
track of the recourse to that method, so that the outcome of the refinement
cannot be reproduced from the deposited data in an automated environment
such as the EDS server.
This may not be the case with the structure Tim was mentioning, but it
provides an opportunity to point out that the use of some recent approaches
to improving refinement at low resolution may need further attention at the
deposition stage to ensure reproducibility of the results from the deposited
information.
With best wishes,
Gerard.
