I disagree with Herman that 3 sigma features are "nothing to worry
about". Indeed, I think this is one of the deepest and most
disturbing problems with macromolecular crystallography in
general. Why can't we explain our data to within experimental
error?
If you think all the green and red stuff left over at the end of
refinement is "just noise", then I challenge you to re-collect
your dataset and do the refinement again. You will find that the
features "wiggle" a bit, but the difference peaks consistently pop
up in the same place. "Noise" doesn't do that.
But, if the only thing you are worried about is "validation" (aka
"is my structure worse than average?"), then yes, once you get to
the point when you can't find any way to build into the tallest
difference feature you've got without making your R/Rfree worse,
then you have reached the limits of current modelling technology
and are basically done with building and refinement. Might as
well deposit what you've got and wait for some future breakthrough
to finally come up with an Fcalc that can explain your Fobs to
within sigma(Fobs).
-James Holton
MAD Scientist
On 6/26/2013 12:08 AM,
[log in to unmask] wrote:
[log in to unmask]"
type="cite">
Dear Jiang
Yan,
The Matthews
function is based on an average protein crystal with 50%
solvent. However, crystals do exist with as little as 25%
solvent or as much as 75% solvent, so if your structure
refines to an Rfree of 20%, your structure is solved and you
have a crystal with a high percentage of solvent (which may
explain your low Rfree). Maps are usually contoured at sigma
levels, based on the variation in the electron density map.
So with an Rfree of 20%, your difference map will be very
flat with and the sigma will be very low. Also the 3 sigma
level usually used for difference maps will be very low and
statistically, there are always some peaks at plus or minus
3 sigma, but they are nothing to worry about.
Best,
Herman
Hi everyone,
I now meet some problems when trying to solve
structure.Space group is P6422, and Mathews
function shows there are 4 molecules in one
asymmetry
unit. However, Phenix-autobuild shows only 2 molecules in on
one asymmetry unit, after refinement,
Rfree=20%,(resolution is 2.8A). Although AA fit the map
well, there are still some green and red density.
I
wonder if anyone met this problem before? Any suggestion
is welcome.
Best,
Jiang
Yan
