An Fo-Fc map is actually the real-space representation of the Fo-vs-Fc R
factor (Rcryst), so the "sigma" of this map will continue to drop
relative to the "true" electron scale as your model improves and the
difference between Fo and Fc diminishes. The 2Fo-Fc map, however, is a
best guess of the "total" electron density, and that is always on about
the same scale, so the "sigma" of this map is pretty constant throughout
refinement. Toward the end of refinement, when your R/Rfee are around
20%, the "sigma" of the Fo-Fc map will be about 1/5 of the 2Fo-Fc
"sigma" level. So, eventually you will see things at the "3 sigma"
level in the Fo-Fc map that are not "visible" at the "1 sigma" contour
level of a 2Fo-Fc map. This does not mean they are "noise". I'm not
sure where that "rule" came from. The "noise level" is actually another
1/5th below the Fo-Fc "sigma" level, or 25-fold smaller than the "1
sigma" contour of the 2Fo-Fc map. This is because the "noise" is not
really related to R/Rfree but rather the actual error in the data. This
is roughly half the value of redundancy-corrected Rmerge-like values
such as Rpim. The factor of two is because these latter R statistics
are for intensities and the maps are calculated from amplitudes.
The situation where the "sigma" of the Fo-Fc map is the "noise level"
really only arises when the difference between Fo and Fc (aka Rcryst) is
comparable to Rpim, and that pretty much only happens for small
molecules where all the atoms in the unit cell can be named and
accounted for. Macromolecular models generally don't achieve that. Not
yet anyway.
So, I encourage you to build into Fo-Fc density if you can. Remember,
there is always "something" there, the question is what that "something"
is. And also "what else could it be"? The "bunch of waters" model is
always an alternative hypothesis. It is fairly easy to show that one
model fits better than another, but to show that the difference is
"significant" requires error bars, and that's why we developed the RAPID
procedure:
http://dx.doi.org/10.1073/pnas.1302823110
-James Holton
MAD Scientist
On 3/10/2014 1:41 AM, herman.schreu...@sanofi.com wrote:
Dear Amlan,
The sigma of an Fo-Fc map map depends on the residual noise in your
map. In a well-refined structure, the sigma will be low, so at 3 sigma
it will show very weak features.
My guess is that your ligand is present in partial occupancy and that
you will find it in your 2Fo-Fc map when you scroll down your contour
level. If you see convincing Fo-Fc density without a ligand being
fitted, the presence of the ligand must be real and you can fit it.
However, I would refine a group occupancy for your ligand.
Best,
Herman
*Von:*CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] *Im Auftrag
von *Amlan Roychowdhury
*Gesendet:* Montag, 10. März 2014 09:09
*An:* CCP4BB@JISCMAIL.AC.UK
*Betreff:* [ccp4bb] regarding Fo-Fc map in coot
Dear All,
Some times during model building in coot we have found that at the
position of ligand molecules and water, there is a good Fo-Fc map
(above 3 sigma), devoid of any 2Fo-Fc map.
1.What does it physically mean and why the 2Fo-Fc map was not
generated properly?
2. Can we fit ligand molecule there?
Thanks in advance.
Best Wishes
Amlan.
--
Amlan Roychowdhury.
Senior Research Fellow.
Protein Crystallography Lab.
Dept. of Biotechnology,
IIT Kharagpur.
Kharagpur 721302
West Bengal.
India.
