Dear Dean,
I have not done any systematic studies, but I have struggled quite a bit
with disordered regions and I can give you some practical tips:
The first thing to remember is that when you have true disorder (in
contrast to fitting errors), if you do things right (not introducing
exessive model bias), your density never will be great and the best you
can do is a least bad fit. What I do in these cases is first delete the
disordered residues, and then do a fair number if refinement cycles to
get rid of model bias. Then I rebuild the deleted residues from both
ends of the gap, one or two residues at a time, with further refinement
cycles in between.
If at least the main-chain density stays continuous, I continue
building, as soon as it gets broken I stop. I find that below a contour
level 0.6 sigma one is looking at noise and fitting is generally not
successful. However from 0.6 sigma upwards, fitting is usually
succesful. It depends, of course, on the type of disorder. If there are
a few discrete conformations, density may look very good, even at lower
contour levels. However, if the loop is moving all over the place,
density will be very bad.
So for me, the measure of completion is when the density gets broken
and/or, the levels drops below 0.6 sigma. At this point, there will
still be many isolated blobs of density visible, which clearly must
belong to disordered loops, but I am afraid that is something you will
have to live with. One can spend forever unsuccessfully trying to fit
these. If there is a paper of someone who did a systematic study on a
measure of completetion, I would be interested as well.
My two cents,
Herman
________________________________
From: CCP4 bulletin board [mailto:[email protected]] On
Behalf Of Dean Derbyshire
Sent: Monday, March 11, 2013 2:16 PM
To: [email protected]
Subject: [ccp4bb] Is there a way of assessing how much of a
structure is 'modelable'?
Hi all,
I'm having great trouble building/refining a structure which has
several regions of obvious flexibility/disorder...
Regions where the density is fragmented and I have no clear way
of building the chain (at least with any confidence). The protein is
dimeric with some variation in the extent of those regions which are not
interpretable, and simply copying from the ncs mate does not work!
So the question is how far can one refine such a structure - is
there a measure of 'completion''?
I recall that there is, but I must be getting old. Am I
miss-remembering?
Cheers in advance
Dean
