I actually use the word "noise" a lot, but I try to reserve it for
things that change every time you measure them. A "noise peak" in
electron density would be one that changes if you re-collect the data.
"Noise" should also go away if you average enough data together. I
don't expect that is the case here, however, so yes I agree that this
should not be called a "noise peak".
Undesirable density that is reproducible and refuses to go away with
averaging is what I call "systematic error": a problem with the model.
Water occupancies are a tricky thing because of that pesky bulk solvent
(which Dale invented, darn him! ;). The electron density of pure water
is 0.335 e-/A^3, and the height of an oxygen (aka "water") with B=30 is
1.5 e-/A^3. So a "water" at 25% occupancy stands only as tall as the
disordered bulk solvent. Now, since adding a water (even at 1%
occupancy) excludes bulk solvent you can actually remove electrons from
a region by adding low-occupancy waters to it. This is important to
bear in mind!
By the way, water occupancies can realistically sum to 125%. This is
because the water molecule has 10 electrons in it, but we always model
it with a bare oxygen (8 electrons). The hydrogens may not be
"visible", but the electrons are there.
Oh yes, and the bulk solvent can be denser that pure water. The formula is:
rho (e-/A^3) = conc*Z*6.022e23*(10/1e10)**3
so 1 M Cl ion (conc=1; Z=18) will add 0.011 e-/A^3 and subtract about 1
M of H20 (Z=10) for a net 0.005 e-/A^3. Not too much difference.
Also, when doing occupancy refinement it is important to always try
multiple starting conditions. That is: start the atoms of interest at
occ=1.00 and see where they go. Do more than 3 cycles of refinement!
Sometimes occupancies converge slowly. Then set them all to occ=0.01
and see where they go. Set them to random values and see where they
go. etc. In this way you can at least get an "error bar". Once you
know what the slop is, you can think about what physically and
chemically reasonable model might fit within it.
I also like to try using a "water" atom that is too big. This is mostly
because occupancy banging up against 1.00 tends to make your error bar
smaller than it should be. Phenix lets you change the maximum allowed
occupancy, but refmac does not. Either way, at B>20 or so all atoms
look the same: they are Gaussians. The only difference between elements
(besides geometry) is the number of electrons under that Gaussian. The
best way to integrate the number of electrons in an unknown, noisy peak
is to fit a Gaussian to it and then take the integral of the curve. The
nice thing about occupancy is that all you need to do is multiply it by
the atomic number and you've got the integral of the electron density.
Very convenient. Yes, occupancy and B factor are correlated, but this
is only because they both impact peak heights. If you really aren't
sure what a blob is, then let both vary and see what multi-start tells
you about how well you can measure the integral.
Underneath all this, however, is the bulk solvent, which contributes
invisible electrons into some nooks an crannies, but not others. Exactly
how much depends on the local geometry, on the bulk solvent parameters,
and on the choice of refinement program. Most default settings have a
"shrink radius" about half the size to the "solvent radius", which
effectively means the edge of the bulk solvent cloud comes right up to
the edge of the atoms on the surface of your protein. For high
resolution structures the "surface" is usually the ordered water shell
itself, so this tends to connect the two water models (ordered & bulk)
together. Nevertheless, adding a new water atom to the coordinates adds
another layer of exclusion, and that pushes the bulk back. This is
where we start getting into trouble. The second shell of ordered water
is no doubt "floating" on top of the bulk, but if we build in a second
layer of low-occupancy waters it prevents the bulk from reaching the
first layer. Refmac has a feature (solvent exclude DUM) that allows
"DUM" atom types to be invisible to the bulk solvent calculation, which
can help, but in general these models are hard to build.
For all these reasons, I recommend having a look at the calculated bulk
solvent density. In refmac5 you can simply specify "MSKOUT solvent.map"
and then load the resulting map in coot. This mask will have no
B-factor applied, however, so you can't see how the density "bleeds"
into surrounding areas. If you want to do that, use SFALL to convert
the map to structure factors. You can load those in coot and then use
the map sharpening/blurring feature to apply the best-fit solvent B
factor. You can find this best-fit B factor in the refmac log under the
scaling step:
Partial structure 1: scale = 0.3751, B = 28.4115
the default "Partial structure" is the bulk solvent. the "scale" is the
electron density of the bulk in e-/A^3.
For phenix you will need to run phenix.fmodel twice: once with the bulk
solvent turned on and once with it turned off. You will want to look
through your phenix refinement log for the optimal "solvent radius" and
"shrink truncation radius" values. Provide these to phenix.fmodel as
mask.solvent_radius=? and mask.shrink_truncation_radius=? . Ideally you
also want the best-fit k_sol and B_sol values as well, but since
phenix.refine no longer uses k_sol nor B_sol, I'd say run it with
k_sol=0.4 and B_sol=<average atomic B>. Once you have your two sets of
calculated structure factors you need to subtract them. You can do this
with the CCP4 program sftools, you can calculate *.map files with FFT
and subtract them with "mapmask", or you can load both mtz files into
coot and make a difference map there using map extensions. All these
operations should be equivalent.
Once you have the map up you can finally ask the question: "is the bulk
solvent involved in that green blob?". If it is not, then maybe it
should be. You can try to let phenix optimize the bulk solvent
parameters for you with
optimize_mask=true
In refmac this is:
solvent optimise
Note the spelling difference, it matters. You can also adjust the bulk
yourself. In refmac this is keyword:
solvent vdwprobe 1.2 ionprobe 0.8 rshrink 0.8
in phenix, they are:
mask.solvent_radius=1.11 mask.shrink_truncation_radius=0.8
Yes, the defaults are different, and refmac has one more parameter: a
different radius for ionizable groups. This may be important in
Jessica's case! But, as you might expect, it can be hard to "fix" a
green peak in one part of the map without creating red peaks elsewhere.
You only have 2-3 numbers to play with and they impact every nook and
cranny in your model. If bulk can't be made to explain the difference
peak, then try modelling a water into that peak, and then look at the
bulk solvent mask again.
With a little work, you can usually come up with a model that both makes
geometrical sense and explains the density.
-James Holton
MAD Scientist
On 3/5/2020 9:50 PM, Dale Tronrud wrote:
On 3/5/2020 11:00 AM, Jessica Besaw wrote:
Hello Matthias,
Excellent point. Most of the the ordered water are easily visible at 2
rmsd. The central disordered (or partially occupied) water becomes
visible only at 1.3 - 1.4 rmsd, and it is very visible at 1 rmsd (which
I have displayed all of the maps). In your opinion, do you think this
would be noise?
"Noise" is a word that I hate. Usually it is just a label we put on
something that we want an excuse to ignore. If you decide to ignore
something you should be honest and have a specific reason.
Personally, I think it is sufficient reason to not place an atom when
you are unsure if an atom should be placed. With the density you have
shown, it is hard to imagine a consistent model that contains a
partially occupied water molecule in this little peak. That "water
molecule" would be inconsistent with full occupancy of the atom to the
left, but that atom already has the lowest B factor when refined at full
occupancy. These two atoms are too close together to both be present in
the same unit cell.
Without a reasonable atomic model in mind, you can't build a
reasonable model. Building a model is not simply the task of placing
atoms in peaks. The resulting structure has to make physical sense.
What hydrogen bonding network are you proposing to exist when this rare
conformation is present?
If you don't have confidence in a model, don't build it. You will be
left with this little difference map peak, and you will take a tiny hit
on your R values. Them's the breaks! Putting in an atom you don't
believe would give you those slightly smaller R values, but is that
honest? It seems to me that building a model that doesn't make
chemical sense just to lower R values borders on deception.
Whether that makes this peak "noise" is irrelevant. Maybe in five or
ten years someone will come up with a new tool for building overlapping,
partially occupied, water networks and this peak will be explained.
Would that change if this peak is "noise" or not? All we have now is a
peak that we don't have a good way to interpret.
Dale Tronrud
Jessica
On Wed, 4 Mar 2020 at 12:45, Barone, Matthias <[email protected]
<mailto:[email protected]>> wrote:
hey Jessica
a tip that might come up later on anyway: once you put every
reasonable bit into the desity, what I like to to when facing such
blobbs: I take a well defined water out to create a diff density at
a position where I know it is real. Having a feeling of how much you
have to contour the diff density at that point can give you a good
feeling how much of noise is actually in your density right in
between the waters..?
best, matthias
Dr. Matthias Barone
AG Kuehne, Rational Drug Design
Leibniz-Forschungsinstitut für Molekulare Pharmakologie (FMP)
Robert-Rössle-Strasse 10
13125 Berlin
Germany
Phone: +49 (0)30 94793-284
------------------------------------------------------------------------
*From:* CCP4 bulletin board <[email protected]
<mailto:[email protected]>> on behalf of Jessica Besaw
<[email protected] <mailto:[email protected]>>
*Sent:* Wednesday, March 4, 2020 6:42:34 PM
*To:* [email protected] <mailto:[email protected]>
*Subject:* Re: [ccp4bb] A question of density
Hey Nukri,
Here are the details: Rwork/Rfree = 0.21 / 0.23 for a 2 Angstrom
structure
I absolutely agree with you on the refinement. I did previously do
that, and I attached the picture.
What is the BB?
Cheers!
Jessica
On Wed, 4 Mar 2020 at 12:20, Nukri Sanishvili <[email protected]
<mailto:[email protected]>> wrote:
Hi Jessica,
You do not say how well is the rest of the structure refined.
First, you should refine the structure best you can, without
placing anything in the unclear blob of your interest so to
obtain the best possible phases and hopefully improve the blob
density as well.
Then you should let the BB see what that density looks like.
Looking at only the list of possibilities has very little value
without seeing the density itself.
Best wishes,
Nukri
On Wed, Mar 4, 2020 at 11:10 AM Jessica Besaw
<[email protected] <mailto:[email protected]>> wrote:
Hello friends,
I have a "blob" of density in an active site of my protein.
I am struggling to determine if I should place a water in
this spot, if I should model it as a disordered water, if
the density may be a ligand that I have not considered, or
if it should be left as unaccounted for density. I would
like to publish this structure without compromising the science.
I have attached several possibilities that I have considered
below.
Any suggestions would be appreciated.
Cheers!
Jessica Besaw
------------------------------------------------------------------------
To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
------------------------------------------------------------------------
To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
------------------------------------------------------------------------
To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
########################################################################
To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
########################################################################
To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
