Simple test is to vary the occupancy (say increments of 0.1) and check for
residual densities following quickie refinements on each. Then at least you
know if you can make a conclusion or not.
I suppose you could also try refining coupled occupancies on ligand-sized
chunks of the protein to get a sense of the reliability of a joint refinement -
not sure if anyone ever tried this..
I am pretty sure you can detect small occupancy errors on occasion - once
with 2A data I had a half-occupied (since it was very close to a symmetry
mate across a 2-fold) Na atom inadvertantly modeled as water. The B-factor
shrank to significantly less than any other atom in the structure until we
replaced the water O atom with Na (and yes, we pretty much know it was Na
because it was replaceable by other monovalent cations and was making
several 2.4A interactions).
I'm not sure about these arguments indicating very high correlations with B-
factor. Of course, occupancy impacts a scattering factor equally across the
entire resolution range but the impact of a reasonable B-factor is significant
only for quite high resolution. Varying B across a reasonable range won't
change scattering much except at high resolution. So the two parameters
become increasingly decoupled at lower resolutions and it is the occupancy
that is the more refinable parameter! For this reason I think it is quite
reasonable to do occupany searches with largish entities like ligands.
|