I have to agree with Ed Pozharski here. It has been shown that it can be
valid to use I/sigma levels as low as ~1 for refinement (Ling, Read, et
al, Biochemistry 1998; Delabarre, Brunger, Acta Cryst D, 2006). I am
bothered more when I see I/sigma cutoffs of >4, where Rsym is <30% in
the high resolution bin. It makes me think the authors might be hiding
something, or stuck with the ancient notions of a not-to-be-exceeded,
sacred Rsym value. Just because the reader might not read the statistics
table does not mean legitimate data should be discarded during refinement.
At the end, it is what inferences you make from your model that
determine your claim of resolution limit (2.5 or 2.6 A!) to be much
relevant. And I do agree with not making too much of the resolution
limit, and presenting your statistics plainly and clearly in a table
(probably not buried in supplementary table 3).
Engin
P.S. Oh well, the thread is hijacked now.
-------- Original Message --------
Subject: Re: [ccp4bb] FW: pdb-l: Retraction of 12 Structures....
Date: Fri, 11 Dec 2009 16:58:07 -0500
From: Christopher Bahl <[email protected]>
Reply-To: Christopher Bahl <[email protected]>
To: [email protected]
I think that when a model's resolution is clearly stated in a paper,
many readers still assume the pre-maximum likelihood definition (i.e.
high I/sigma, low Rsym in the high resolution shell). I've never seen a
paper where the I/sigma was given in the abstract after stating a
resolution. This can potentially mislead the average reader's
perception of the "actual" resolution (if it exists). It is my opinion
that authors should not proclaim a resolution for their structure if
they aren't employing the same stringency that has classically guided
the limits of resolution. Just leave that sentence out and let the
statistics table do the talking.
-Chris
--
Christopher Bahl
Department of Biochemistry
Dartmouth Medical School
7200 Vail Building, Rm 408
Hanover, NH 03755-3844 USA
phone: (603) 650-1164
fax: (603) 650-1128
e-mail: [email protected]
Ed Pozharski wrote:
Not to derail the thread, but there is nothing, imho, wrong with I/s=1
cutoff (you expect I/s=2, I assume?). R-factors will get higher, but
there are good reasons to believe that model will actually be better.
This has been discussed many times before and there is probably no
resolution, so why not just let people choose whatever resolution cutoff
they want (as long as the I/s is clearly stated)?
Disclaimer: I always use I/s=1 cutoff (assuming that completeness is
good, of course). Compared to I/s=2 it doesn't really overstate
resolution all that much (e.g. 2.1 vs 2.2).
On Fri, 2009-12-11 at 13:18 +0100, Silvia Onesti wrote:
I think also the editors are sometimes to blame.
I once refereed a paper and pointed out that the resolution was overstated
(I/s(I) = 1.05 in the last resolution shell, as well as a couple of comments
that clearly suggested that the density wasn't very good). The editor
ignored my comments.
Silvia
------------------------------------------------------------------------------------------------
Silvia Onesti
Sincrotrone Trieste S.C.p.A.
SS 14 - km 163,5 - AREA Science Park, 34149 Basovizza, Trieste ITALY
Email: [email protected]
Tel. +39 040 3758451
Mob +39 366 6878001
http://www.elettra.trieste.it/PEOPLE/index.php?n=SilviaOnesti.HomePage
http://www.sissa.it/sbp/web_2008/research_structuralbio.html
------------------------------------------------------------------------------------------------
On Fri, 11 Dec 2009 10:48:41 +0100
Vellieux Frederic<[email protected]> wrote:
Hi all,
Like everyone else, I was appalled.
My two cents worth: Nature and Science are not scientific journals in the
strict sense of the term. They are more like magazines (I won't go all the
way
to say "tabloids"), and as such will do anything to publish what seems to be
hot. And will reject very good scientific papers. So it's not a surprise
that
retractions affect magazines such as Science and Nature.
Fred.
