Transfer of Rfree flags to a new data set
There are many cases in which it is appropriate to replicate
the set of reserved reflections used to track Rfree so that the
same set is used for multiple data sets. Some examples of this
are:
-
You start refinement of your structure, but become worried
that there was an error in the data processing.
You re-run denzo/scalepack/scala/truncate or whatever to
produce a new reflection file. The first thing you might
want to test is whether your re-processed data is in fact
better. If it is, then you probably want to continue your
refinement from where you left off but using the new data
file. In either case you should keep the same set of
reserved reflections to track Rfree.
-
You start refinement of your structure against one data set,
and then while your refinement is in progress you get a better
data set (higher resolution, more complete, whatever).
It would not be correct to select a new set of reserved reflections
from your new data set randomly. They would no longer be "free".
They would, in fact, be replicate measurements of reflections that you
already used in refinement of your first data set.
Instead you should reserve the same set of reflections in your new
data set as in your old one. If your new data set goes to higher
resolution, then you augment the original set by adding
reserved reflections from your new high-resolution shell of data.
-
You are refining highly isomorphous structures, for instance multiple
parallel soaks of different small molecules into the same bunch of
crystals. You will presumably begin each structure refinement/analysis
from the same starting model. Neither the models nor the data are
independent, so you should not allow reflections used in the refinement
of the starting model to be treated as "free" in the later refinements.
Instead you should use the same original set of reflections for Rfree
in each case.
Imperfectly isomorphous structures present more of a borderline case than
the examples above.
Just how isomorphous do two structures have to be before the data sets are
not independent? It's something of a judgement call. But on the other
hand, it doesn't hurt to keep the same set of reserved reflections
so you might as well do it anyway.
I now give a pair of fragments from X-PLOR scripts that may be used to extract
the Rfree flags (TEST) from an existing data file and merge them back into
another data file.
!
! Read in an existing data file,
! write out only the Rfree flags
!
@@crystal_setup.inp
nreflections = 50000
reflection @olddata.xpl end
write reflection
test
output= oldtest.xpl
end
stop
!
! Merge Fo, Sigma from new data set with
! Rfree flags (TEST) from previous data set
! Write out merged data set to newdata.xpl
!
@@crystal_setup.inp
nreflection = 100000
reflection @newdata.fob end
reflection @oldtest.xpl end
resolution 100. 2.0
reduce
write reflection
fobs sigma test
output=newdata.xpl
end
mbins = 20
print completeness
?
stop
E A Merritt 1999