On 21/01/13 14:21, Clemens Vonrhein wrote:
Dear all,
On Thu, Jan 17, 2013 at 07:48:34AM -0800, Nat Echols wrote:
If I had to guess, I'd say that the output from Refmac is on an
approximately absolute scale, i.e. the volume-scaled density values
resemble the actual electron densities expected for the model. (I say
"approximately" because the absence of F(0,0,0) and various FFT
artifacts pretty much guarantee that the values are not actually
accurate, just on the same order of magnitude.) This is presumably
done by scaling F(obs) to F(calc). Maps from Phenix are definitely
*not* on an absolute scale, however, and I guess the same must be true
for BUSTER.
Just for the record: the map-coefficients written by BUSTER
(amplitudes 2FOFCWT and FOFCWT in the final 'refine.mtz' file) are on
the same scale as the model (amplitude FC) ie. on approximate absolute
scale.
As far as we can tell, Coot uses the actual map-values (e/A^3) as a
'score' in those 'density fit' graphs - is that right?
Yes - for the moment at least. If I can make it fast enough, it will
not be the case for 0.8.
So it is not
something with a known range (like a real-space correlation value from
-1 to +1) - especially not if the maps are sometimes on roughly
absolute scale and sometimes they aren't.
Yes.
It seems as if the determination of apropriate range for drawing those
bars within coot is affecting if one sees only red (false negative) or
only green (false positive) bars.
Yes.
Maybe Paul/Kevin/Bernhard can comment on how this is done? How does
one avoid getting frustrated by too many red bars and (equally
important) over-excited by bogus greenery?
No good answer at the moment. The graph is only relative :-( I can
only refer you to Bernhard's answer (above).
Paul.
