*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
Okay, this discussion has gone beyond Rmerge, but it does inspire an old
question of mine that I have never been able to find a good answer to:
Does zero-filling help crystallographic refinement?
Our NMR colleagues have been doing this as standard practice since the
1980s. They "pad" missing data off the end of the FID acquisition with
zeroes because they have the prior knowledge that the signal does, in
fact, end up being zero out there. It stands to reason that
crystallographic refinement might benefit from the practice of forcing
high-resolution hkls to be zero. Otherwise, errors in the model will
tend to get "squeezed out" into these high-order terms (a general
consequence of having more parameters than observations). If you look
at a calculated Wilson plot from some old entries in the PDB, you will
see Fc drop to near zero at the reported "resolution limit", but if you
calculate Fc out beyond the reported "resolution limit" then Fc starts
to become non-zero again.
I once tried filling the outer shells of a data set with zeroes and fed
it to refmac. This caused refmac (4.0) to crash, and I never did get
back to that project.
Garib?
-James Holton
MAD Scientist
Vaheh Oganesyan wrote:
The intensity of the reflection (hkl) being zero is a piece of
information, provided there are other reflections in that resolution
shell that are not zero (above sigma). If all of the reflections in
the resolution shell are essentially zero (<< sigma), then the
resolution cut-off is too optimistic.
/_There are two somewhat separate questions here: (1) what data to
use in refinement and (2) how to define a "nominal resolution"
_/
How do you see those questions being separate? Let say you collected
data to resolution A (edge or corner of detector), then refined to
resolution B. You probably looked at the data and considered that the
data beyond B are either compromised by experimental setup or just
absent. Why would anyone consider calling the resolution of the
refined model or resolution of the presented structure other than B?
One may state in the main body of the paper that the crystal actually
diffracted to much higher resolution A but for such and such reasons
we were able to use the data only till resolution B.
that will give the outside world a good idea of how far out the data
extended, which needs some vague sort of uniform standard. I'd
suggest the answers are (1) any datum you believe you really did
measure should be thrown into the pot and (2) the reported nominal
resolution should be where I/sigI falls below 2, perhaps with the
output statistics from scalepack or whatever was used included as
supplementary material to show that the sigmas are not grossly
mis-estimated. Unfortunately this creates logistical problems with
deposting coordinates, etc. if you want to call the official
"resolution" something other than that of the farthest-out fly speck
you refined against.
At 09:29 AM 4/28/2006, Bart Hazes wrote:
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
This is an interesting conceptual question. A reflection with an
intensity of virually zero does not contribute to your electron
density map, so can it be data? If it is can I claim to have a 0.02
Angstrom structure although all intensities beyond 2 Angstrom are
virtually zero?
The answer to question 1 is YES it can be data
The answer to question 2 is NO KEEP DREAMING
A very weak reflection does not contribute to your map so leaving it
out doesn't hurt since missing a reflection is equivalent to
assuming it is zero, not a bad approximation. However, a lot of our
maps are of the 2Fo-Fc and Fo-Fc types. Similarly refinement a
depends on |Fobs-Fcalc|. Thus, if Fcalc is big and you measured it
as being near zero then the discrepancy is large and the reflection
has a large contribution to refinement and these kinds of maps. Had
you not measured the reflection then its contribution would be
ignored (equivalent to Fo-Fc being zero).
The difference at the high resolution limit is that Fcalc starts to
approach zero as well as Fobs. So now both Fobs and Fo-Fc are
virtually zero and small relative to the errors in Fo and Fc. So
this time there really is no information left to speak of.
Bart
[EMAIL PROTECTED] wrote:
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
What is "observable data"? If I observe a reflection by passing it
through
the Ewald sphere I can measure its intensity
Phil
Here's what Acta Cryst has to say about resolution-
From the "Notes for authors 2006" from Acta Cryst D:
http://journals.iucr.org/d/issues/2006/02/00/me0308/index.html
"11.1. Resolution
The effective resolution should be described clearly. Values of the
internal
agreement of the data, Rmerge, together with the multiplicity, the
mean
value
of I/ and the percentage completeness of the data are required for
the
overall
data set and the highest resolution shell together with the limits
of that
shell in Å. For high-quality data obtained with synchrotron
radiation,
values
of Rmerge < 20%, completeness > 93% and observable data >
70% should
be
achievable for the highest resolution shell. A complete table
listing the
above
criteria as a function of resolution should also be submitted, but
will
normally
be included in the supplementary material..."
---------------------------------------------------------------------------------------------------------------------------
Phoebe A. Rice
Assoc. Prof., Dept. of Biochemistry & Molecular Biology
The University of Chicago
phone 773 834 1723
fax 773 702 0439
http://bmb.bsd.uchicago.edu/index.html
http://www.nasa.gov/mission_pages/cassini/multimedia/pia06064.html
