I am a bit out of touch with the discussion, and this may have been
mentioned already.
It is important to remember that Sigma is an OVERALL value for the whole
map, whereas one is looking for local solutions when fitting any
density. Stuff on the surface of the molecule ought to be contoured at a
lower level than in the ore, and this applies to protein as well as waters.
Eleanor
Ed Pozharski wrote:
On Wed, 2010-04-21 at 17:21 -0700, James Holton wrote:
The "0.3% chance" of a peak being above 3
"sigmas" assumes that the histogram of electron density values is
Gaussian. It is not! In fact, it is a funny-looking bimodal
distribution (the peaks are protein and solvent regions).
Indeed! That's why it is a "bizarre" argument. In fact, standard
deviation is rather meaningless unless one is dealing with "univariate"
distribution. For bimodal distributions, changes of standard deviation
are uninterpretable (without looking at the distribution, that is) since
they can be due to both shifts and redistribution.
Cheers,
Ed.
