If you are seeing "waters" at 4 A you are probably looking at ions.
Those are OK to model, but definitely try occupancy refinement from a
variety of starting points to make sure you are not fooling yourself.
Occupancy refinement might also help you assign which ion it is. That,
and some prior knowledge about what ions are in your buffers.
As for "random noise", if you re-collect the data and re-refine your
model, and find that the feature is still there, then it is NOT random
noise. Random noise (by definition) changes every time you measure
something.
As for Fourier ripples, there is an easy check for this: 1) calculate
structure factors from your refined model to 1.0 A or so (even if you
have 4A data). 2) flag all your observed hkls, set the FC of those hkls
to zero (use SFTOOLS for this). 3) use the remaining FC values to
calculate an electron density map. This map will be the difference
between a "perfect" map (essentially no missing reflections), and one
that has the same missing Fourier terms as your observed map. This is
the best estimation you can make of where you expect to see Fourier
ripples, given your particular model and resolution cutoff. You will
find that the ripples are generally very very small. Unless your
observations cut out way too much good data.
If you look at a calculated map at 4A with waters in it, such as found
in my movie here:
http://bl831.als.lbl.gov/~jamesh/movies/
You will find then you will find that the waters disappear below 1 sigma
at around 2.8A resolution. It doesn't mean they don't exist, they just
drop below the 1 sigma contour level. If you lower the contour you will
see them again. At what point does lowering the contour level get you
to "noise peaks"? Well, that's around -1.0 sigma. Yes, negative
sigmas. Everything above that is "real" (Lang et al. 2014
http://dx.doi.org/10.1073/pnas.1302823110). Just difficult to build into.
-James Holton
MAD Scientist
On 4/13/2015 12:12 PM, Phoebe A. Rice wrote:
At 4A, I wouldn't unless I had an exceptionally good reason to.
There will always be some blobs, due to random noise and fourier
ripples as well as due to an imperfect model. Unless a blob makes
nice H-bonds to something else that is nicely ordered, I wouldn't
model at water into it. If you can't see nice density for side chains
then you probably aren't really seeing density for waters either.
++++++++++++++++++++++++++++++++++++++++++
Phoebe A. Rice
Dept. of Biochemistry & Molecular Biology
The University of Chicago
pr...@uchicago.edu <mailto:pr...@uchicago.edu>
------------------------------------------------------------------------
*From:* CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of
Sudipta Bhattacharyya [sudiptabhattacharyya.iit...@gmail.com]
*Sent:* Monday, April 13, 2015 1:14 PM
*To:* CCP4BB@JISCMAIL.AC.UK
*Subject:* [ccp4bb] Picking water molecules at 4A structure.
Dear community,
Recently we have been able to solve a crystal structure of a
DNA/protein complex at 4A resolution. After almost the final cycles of
model building and refinement (with R/Rfree of ~ 22/27) we could see
some small water like densities...all throughout the complex. Now my
query is, whether one should pick water molecules at this low
resolutions or it is totally unscientific to do so?
Many thanks in advance...!!!
My best regards,
Sudipta.
