A general point often overlooked, the Rfactor from twinned refinement
at least with SHELX and REFMAC do not use exactly the same formulation
as an Rfactor for a mono-crystal and seems consistently lower than
expected.. my impression is that they cannot be compared easily.
And there may be a problem with the FreeR set which should reflect the
twinning pairs. I belive PHENIX deals with this properly, but it is not
nec so for CCP4. You need to assign the Free sets in the highest
conceivable Laue group then extend them to the lowerr spacegroup.
Eleanor
Garib Murshudov wrote:
I am getting in habit of writing double emails.
I would say that refmac overestimates at early stages and truncate
underestimates (it is just an intuition, not based on theoretical or
empriical results)
Garib
On 23 Apr 2010, at 23:16, Ethan Merritt wrote:
In a nutshell
=============
Is there a way to make solve/resolve behave reasonably if the data
is twinned? Is there a recommended alternative path to clean up and
maybe even auto-trace a map with 4-fold NCS but twinned data?
Maybe Pirate/Buccaneer?
In detail
=========
I'm fighting with a structure that is probably/maybe in P4(2).
2.6A data 98% complete
Rmerge = 0.07 in P4
Rmerge = 0.16 in P422 (note the disturbingly low value)
Pointless is 88% confident it's P4(2).
Laue Group Lklhd NetZc Zc+ Zc- CC CC- Rmeas R-
= 1 P 4/m *** 0.916 4.68 9.36 4.67 0.94 0.47 0.09 0.30
10 P 4/m m m 0.000 6.83 6.83 0.00 0.68 0.00 0.20 0.00
Truncate says the data is twinned, with a twinning fraction of 0.25.
That is consistent with the bad-but-not-awful Rmerge in P422, right?
I have a promising initial MR solution in P4(2) with four monomers
in the a.s.u. It refines to R/Rfree = 0.36/0.38 in refmac but only if I
enable the twinned data option. Twinning fraction refines to
Twin fractions = 0.4322 0.5678
My thought was to use the 4-fold NCS and solvent flattening in resolve to
clean up and ideally auto-trace into the initial mediocre maps.
The resolution is not great, but the 4-fold NCS should help a lot.
Unfortunately, so far as I can figure out resolve ignores the twinning
and is stuck dealing with R > 0.60 and extremely poor maps.
I tried de-twinning the data, but this was not a great success.
Should I believe the truncate estimate of 25% twinning fraction
or the refmac estimate of 43%? Neither? In any case, feeding the
supposedly detwinned data to either refmac or resolve produces
worse results that I was getting before attempted detwinning.
Any advice?
--
Ethan A Merritt
Biomolecular Structure Center
University of Washington, Seattle 98195-7742
