[phenixbb] Very low rmsd bond length
Tim Gruene
tim.gruene at psi.ch
Mon Aug 7 08:57:32 PDT 2017
Dear Luke,
I missed where your quote comes from, so please apologise if I address you
incorrectly.
The statement
We prefer to maintain Rfree and (Rfree-Rwork) both as small as possible to
minimize the possibility of overfitting.
is not quite right: there is an expected gap between Rfree and Rwork, i.e. if
this gap becomes too small, something is wrong, e.g. a real contamination of
Rfree by an oversight of Twinning, NCS, or a higher space group (Ian Tickle's
articles on this topic is probably the best reference and available at Acta
Cryst).
Best,
Tim
On Wednesday, July 26, 2017 1:45:55 AM CEST Luke Rice wrote:
> Mark,
>
> I think a reasonable strategy is to acknowledge the concern, inform the
> reviewer about what changes if you do as asked, and use that as a
> data-driven way of justifying your decision. I pasted below what worked
> for me when I had to address a similar issue in review.
> > “the geometry is too tight”
> > This is a frequently debated point among crystallographers, and there
> > is no clear consensus. Given the challenges of our data (anisotropy,
> > moderate completeness and resolution), we do not believe there is
> > sufficient information content in the data to warrant significant
> > deviations from ideal bond lengths and angles. Test refinements using
> > different weights between the X-ray and stereochemical terms bear this
> > out: allowing the geometry to relax by choosing a weight such that
> > bond and angle deviations in the refined model are 0.01 Å and 1.3°
> > results in Rfree decreasing by 0.3% while Rwork decreases 2.5%. Thus,
> > we could have achieved a slightly lower Rfree at the expense of
> > significantly greater decrease in Rwork. We prefer to maintain Rfree
> > and (Rfree-Rwork) both as small as possible to minimize the
> > possibility of overfitting.
>
> Good luck,
>
> Luke
>
> > On Jul 25, 2017, at 5:52 PM, Mark Saper <saper at umich.edu> wrote:
> >
> > Hi all,
> >
> > A reviewer points out that my refined structures (which have already been
> > deposited in PDB) have very small RMSD values for bond lengths
> > (0.005–0.002) (refinement resolutions are 2.0–2.8 Å), and that my weights
> > should be decreased. The weights were calculated with “Optimize XYZ
> > Weight” option turned ON. This seems to be the case with all of the
> > structures that I have refined during the past 5 years. My impression
> > from previous discussions is that deriving explicit weights from these
> > optimized ones is not straightforward and so I have never tried to relax
> > the weights. Moreover inspecting the optimization results in the log
> > file, suggest that changing the weights to increase the RMSD would
> > increase the Rfree. (I can post this if its helpful).
> >
> > How do I address the reviewer’s concerns?
> >
> > Thanks,
> > Mark
> > _______________________________________________
> > Mark A. Saper, Ph.D.
> > Associate Professor of Biological Chemistry, U-M Medical School
> > Room 3220B, MSRB III saper at umich.edu +1 (734) 764-3353
> >
> >
> > _______________________________________________
> > phenixbb mailing list
> > phenixbb at phenix-online.org
> > http://phenix-online.org/mailman/listinfo/phenixbb
> > Unsubscribe: phenixbb-leave at phenix-online.org
>
> ________________________________
>
> UT Southwestern
>
>
> Medical Center
>
>
>
> The future of medicine, today.
>
>
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb
> Unsubscribe: phenixbb-leave at phenix-online.org
--
--
Paul Scherrer Institut
Tim Gruene
- persoenlich -
OFLC/104
CH-5232 Villigen PSI
phone: +41 (0)56 310 5297
GPG Key ID = A46BEE1A
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: This is a digitally signed message part.
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20170807/d8bc63f1/attachment.sig>
More information about the phenixbb
mailing list