Something you want to be very careful with here that appears to have not
been mentioned yet is "phase bias".
If you have built in a wrong side chain or other entity that is not
actually there, but still force it to have occupancy=1 and reasonable B
factors (B factor restraints might do that) and then you refine the rest
of the molecule to convergence, build in lots of waters, etc. it is
quite astonishing how much "memory" the resulting phases can have for
the incorrectly-placed atoms. If you simply remove the wrong side chain
atoms and then re-calculate the map with no refinement, then it is not
unexpected that you will see "green" difference density for it. This
may persist even after a few cycles of refinement.
Best thing to do after removing a "questioned" feature is to jiggle the
rest of the molecule a bit, remove all the waters, and then re-refine to
convergence. If the difference feature survives such a treatment, then
it is probably for real. This used to be called a "simulated annealing
omit map". Something that seems to have been lost in a few modern packages.
And yes, "phase bias" can work the other way as well. If you really do
have a low-occupancy feature, then not modeling it and re-refining
everything else will result in "negative" phase bias: squashing your
weak feature into oblivion. Exactly what occupancy is required to be
resistant to a "simulated annealing omit" procedure? That's a good
question. However, I am reasonably confident that this "critical
occupancy" is less than 1.0 And probably greater than zero as well.
-James Holton
MAD Scientist
On 11/20/2012 10:36 AM, GRANT MILLS wrote:
Hello all,
I'm currently working on a structure which if I stub a certain side
chain phenix/coot shows me a large green blob which looks strikingly
similar to the side chain, when I put it in and run another refinement
the blob turns red.
Basically I was just playing around and I changed the occupancy of the
side chain and now there are no complaints. But I was thinking, should
I haven changed the Bfactors instead? Should I have left well enough
alone? If I lower the occupancy manually and do not include alternate
confirmations have I introduced modelling bias?
Could someone recommend some good articles I could read on exactly how
to correctly fix this problem.
Thanks,
GM
