Sorry for the cliche, but the goal of refinement is not to reduce R
factors, but to produce a good model.
ARP/wARP uses the 'WEIGHT AUTO' option of REFMAC5 to get a good
geometry.
You should set the weight to a value that produces 1-2 and 1-3
distances rms deviations similar to the ARP/wARP job,
to be able to compare. The fact that weight is 0.3 says nothing.
The correct weight can vary wildly from 0.02 to 0.5, in my
experience. for 2.0 data 0.3 sound loose, 0.15-0.2 is what I am used to,
depending on dataset. But, The only way to tell what is right is
inspecting the geometry and aim for a 'reasonable' rms 1-2 distances
deviation.
What is 'reasonable', can cause yet another long discussion, but my
personal favorite for 1-2 distances rms deviation is between
0.015-0.020.
In Refmac these also give the lowest R factors, in my hands.
The invisible side chains is yet another long discussion that you can
retrieve from the ccp4bb archives.
Again, my personal preference is to leave them in and let them get
very high B factors, as long as they do not
get negative density in difference maps, that I presume you are
inspecting.
I dont mind deleting them (but dont like it) and I think mutating to
ALA is worse since its misleading to users.
Finally, given that you have 2.0 A data you should try and model not
only waters, but also:
a. double conformations of side chains
b. solvent and cryoprotectant molecules; glycerol, SO4 etc should be
different than waters and easy to model.
... and I still cant help wondering how people do their phd's or post-
docs in labs that no-one can explain
such trivialities. Or why people prefer not to ask their colleagues
and supervisors, but to mail ccp4bb.
Or why do I bother answering such emails on a Saturday morning, and
then complaining,
only to have the likes of Dr. Walsh commenting about my humor ;-)))
I find all these really scary.
Tassos
On 21 Jul 2007, at 0:29, JOE CRYSTAL wrote:
Dear all,
I am refining a structure at 2.0 A. The water molecules have been
added using arp/warp resulting Rwork/Rfree=21/26% (about 370 HOH
for 360 residues). After 10 cycles of refmac refinement (wt 0.3),
Rwork/Rfree went up about 1.5% to 22.5/27.5%. I did some minor
adjustments and add/delete water in Coot followed by 10 cycles
refmac refinement, but Rwork/Rfree are still around 22.5/27.5%. I
also noticed a few side chains without density. Will setting those
atoms to 0 occupancy or high B factor or mutating to Ala help
decrease Rfree substantially? If not, is there any better
strategies to lower down R factors? I will be very appreciative if
you have any suggestions or comments to offer. Thank you in advance.
Best,
Joe
