From: Eleanor Dodson <ccp4@ysbl.york.ac.uk>
Date: 21 November 2011 10:23
To: CCP4BB@jiscmail.ac.uk
Just a plea for less molecular replacement.
If you get a new crystal of a known protein with the same cell dimension as youur old crystal, the most likely scenario is that it has the same group, and you really should not try MR - use the previous solution as input to do rigid body refinement, and then
a) the R factor will tell you if this is a reasonable hypothesis (it usually is..) and
b) you dont have this awful problem of not being able to compare the solutions..
Eleanor
On 11/20/2011 03:57 PM, Napoleão Valadares wrote:
Thank you all for the replies. Felix Frolow, Dan Leahy, Hans
Brandstetter, Boaz Shaanan and Tim Gruene you really helped a lot.
I think I understand it now, I always thought the "one ring to rule them
all" translated in the crystallography realms to "one origin to rule
them all". That probably means I have a long road in front of me.
I'm still half confused, I definitely need to read more, as much as I
read about symmetry and space groups I never seem to improve or get a
better understanding, but I'll keep trying.
About the same origin:
The pdbs of both Solution-1 and Solution-2 present the same space group
and cell, as observed opening the pdbs as text files or in pymol. When I
open both maps on coot they are not superposed but present the same cell
and origin.
If I open both solutions on pymol they clash. If I generate the symmetry
mates of both solutions none of them are superposed, instead they clash.
But I think they are related as you all pointed, I'll check it out.
Thank you all for your kind answers and your patience with a beginner.
Regards from a sunny Brazil,
Napo
On 11/20/2011 2:58 AM, Felix Frolow wrote:
Napoleao,
It is so called alternative origins play a game with you. You do not
change your structure by shifting 1/2 translation (or even combination
of these translations)
into directions of the main axes of your unit cell. Structure factors
after this operation stay the same, however phases change
systematically, producing however the same
map features.
Would I be a begin crystallographer now, I would read a bit more old
fashioned books
on crystallography such as probably Jensen and Stout…
FF
Dr Felix Frolow
Professor of Structural Biology and Biotechnology
Department of Molecular Microbiology
and Biotechnology
Tel Aviv University 69978, Israel
Acta Crystallographica F, co-editor
On Nov 20, 2011, at 07:42 , Napoleão Valadares wrote:
Hello,
I'm observing a very strange phenomena (at least to me, I'm a
beginner). It is related to symmetry (I think).
I got a data set at 1.95A (I/Sigma 3.5, R-Factor and R-meas < 35% in
the last shell) and a partially refined solution with R/Rfree 22/24,
166 aminoacids observed and around 30 solvent molecules. I'll call
this Solution-1. The refinement was smooth, the densities were very
clearly "asking" for the correct missing side chains and the map
looks good.
The space group I'm using is P212121, pointless and XDS agree with
that (but me and pointless both have a long history of being wrong
about space groups). Phenix.xtriage says there's no twinning.
I took Solution-1 and used it as a template in a molecular
replacement in the same space group (P212121) using the same mtz as
the one used to refine the template. I got a different (not
superposed in space) solution (called Solution-2, scores by Phaser
RFZ=24.2 TFZ=33.0 PAK=0 LLG=1413 LLG=1899) that was readily refined
in Phenix to R/Rfree 24/26 without any solvent molecule.
- The solutions are not superposed in space, although they are
near-identical and can be superposed yielding a C-alpha rmd =0.001.
- Both structures present VERY similar density maps. The maps are not
superposed in space, but when you "run the chain" in one map in Coot
and do the same in the other it they the present exactly the same
features. It is impossible to ignore their similarities.
- Both structures and maps present the same origin and unit cell.
- If I add to Solution-2 the equivalent solvent molecules of
Solution-1 (I did this by superposing Solution-1 to Solution-2 then
copy/pasting the solvent molecules), the R/Rfree become 22/24. This
is a clear indication that the solutions are related.
- I can't find any MR solutions using the same template and space
groups P222, P2221 and P21212.
How two different sets of phases can yield maps with the same
features? What is happening, wrong space group? I have a feeling my
lack of experience is the problem.
Thank you.
Regards,
Napo
----------
From: Ian Tickle
Date: 21 November 2011 11:02
To: CCP4BB@jiscmail.ac.uk
But sometimes (or actually we find quite often) the crystal after
soaking & freezing is sufficiently non-isomorphous (we sometimes see
up tp 10% changes in cell parameters) that RBR just doesn't work, and
then you have to fall back on MR. The solution in that case to avoid
problems of generating symmetry-related and/or origin-shifted
molecules is to do a limited MR search, e.g. rotating/translating each
molecule in the a.u. independently by up to 5 deg & 5 Ang. from the
model (which of course has 100% similarity making it a lot easier).
Furthermore, because the number of points sampled is much less one can
afford to do the more accurate full 6-D search, as opposed to the
usual 3-D RF followed by 3-D TF on each RF solution. So now we do
this routinely (we don't even bother with a preliminary RBR). I
believe the limited 6-D search can be done with Phaser.
Cheers
-- Ian
----------
From: Boaz Shaanan
Or use Kevin's csymmatch which does wonders on scrambled oligomers that (nearly always, at least in my hands) come out of Phaser in cases like those mentioned by Ian and which do need MR.
Boaz
Boaz Shaanan, Ph.D.
Dept. of Life Sciences
Ben-Gurion University of the Negev
Beer-Sheva 84105
Israel
----------
From: Michael Thompson
A question regarding the plea for less MR (which I support):
There have been several recent instances in which I have used the solution of an isomorphous structure to do rigid body refinement for a new crystal (as described by Eleanor). It has always produced good results. My question is about how to best handle the free set of reflections when doing this? I have heard a number of differing opinions about whether or not it is important to carry the freeR flags from the original structure over to the new data set. I have heard equally convincing arguments from both sides, so my young and impressionable mind does not know who to believe. I was hoping I could get an opinion from the "advocates for less MR."
Sorry for hijacking this thread, but hopefully it will provide some insight that is relevant to the original post.
Thanks!
Mike
From: Parthasarathy Sampathkumar
Hi Mike,
Often, I generate independent freeR set (especially in cases where "soak" dataset is of different resolution (usually worse) compared to the native dataset and do following two things to get rid-off the bias: 1. add a noise to the coordinates (this can be done using PDBSET). 2. set the Bvalues to the wilson B of the "soak" dataset (within Refmac5 before rigidbody refinement).
HTH,
Partha
----------
From: Dale Tronrud
I'll jump in here, and avoid the question entirely.
Since running MR on an isomorphous crystal gives the same answer as just
stuffing in the model and running some rigid body refinement, however you decide
to handle your R free flags, you should do the same thing in both cases. The
model is the same.
Dale Tronrud
----------
From: Eleanor Dodson
If you run scala after data integration, ( and thats a good idea whether you have used any integration software - just get out the unmerged scaled file - feed it through pointless to get useful info there, and then scala/truncate to get more useful graphs and analysis) there isa an option to extract the freeR set from a previous data set. That is highly recommended if you have such data.
Eleanor
No comments:
Post a Comment