Thanks Paul,
I don't know of a perfect solution to the question of rms
for an ncs averaged map. Knowing your imperfect solution
clarifies the results I'm seeing and that helps a lot. I
guess I'm seeing greater anomalies than others because I
have 6-fold ncs and 50% solvent. I take it that each copy
of the map ends up being surrounded by a large number of
zeros.
Dale
On 12/22/2010 6:38 PM, Paul Emsley wrote:
When creating an NCS average map, Coot draws a box around the chain of interest
and averages the map inside that. Everywhere else is
set to 0.
So inside the box, yes, one might expect the rms value to be a small amount
larger (considering points inside the box). However,
most of the ASU is set to 0, so overall, the rms of the average map will be
considerably less.
Bottom line is that y.y is difficult to calculate unless you cut the map back
to the box values - I suppose that the output of Coot
and mapmask would help there, although for clarity you would have to explain
how sigma is derived (it is often opaque, it seems to me).
Paul.
On 22/12/10 23:37, Dale Tronrud wrote:
The "12 sigma" I want to put in the paper is from the Fo-Fc map
where the rms is more comparable to a sigma. You are right that
e/A^3 is the best "units" for describing density but for my water
creation criteria I would like to say "higher than x.xx e/A^3 in
the difference map which is y.y sigma."
I would never attempt to ascribe statistical meaning to the
rms of a 2Fo-Fc map. I chose to describe my confusion about
its calculation in Coot for ncs averaged maps because I could
describe my puzzlement more easily. With a Fo-Fc style map you
expect the "sigma" of the peaks to get bigger after averaging
and you have to argue how much larger is plausible. For a
2Fo-Fc style map the density should not get much higher, in
terms of rms, with averaging and yet in Coot it does. My guess
is that the calculation is done the same way for both types of
maps.
Dale
On 12/22/10 15:08, Phil Evans wrote:
Why do you want to quote "sigma" level anyway? It's more or less meaningless
for the reasons you give. Stick to e/A^3
</flame>
Phil
On 22 Dec 2010, at 22:02, Dale Tronrud wrote:
Hi,
I have a crystal structure at 3A resolution with six copies in the
asu. When I average the map over the ncs I find that the original
2Fo-Fc style map has a sigma of 1.5 at 0.3 e/A^3. When I adjust the
contour level of the averaged map to match, by eye, the level of the
unaveraged map I find them equivalent at a sigma of 2.6 at 0.28 e/A^3.
These results imply that the "sigma" level of the original map was
0.2 e/A^3 and the averaged map was 0.11 e/A^3.
The "sigma" of a 2Fo-Fc style map is not an estimate of uncertainty,
of course, because nearly everything in the map is signal. It is
just a measure of the variability of the signal, i.e. the rms. With
averaging the signal should be preserved and the noise reduced, but
the noise of a 2Fo-Fc map is small compared to the signal. How is
it that the rms of my averaged map drops to half of the unaveraged
value and yet the electron density looks about the same when contoured
at the same e/A^3 (0.3 vrs 0.28)?
I guess the real question is, how does Coot calculate the "sigma"
of an averaged map? You can't calculate the rms over the asymmetric
unit because the asymmetric unit is many millions of unit cells in
size (and hugely variable depending on small changes in the ncs
operators).
The problem at hand is that I want to quote the sigma level I
insisted upon when creating water molecules and think it will sound
weird if I say I used a value of 12, which I did. The numbers just
don't seem right to me so I'd like a little reassurance.
Dale Tronrud
