I absolutely agree with Phil.
However I think it is very important NOT to remove systematic absences
before scala. There are many cases of mis-assigned space groups where
"systematic absences" are misleading - due to NC translation or bad
measurement or whatever....
Eleanor
On 01/24/2011 09:32 AM, Phil Evans wrote:
My immediate response to this is that anyone worrying about the number of
reflections should get out more.
More seriously, is assessing the quality of measurements then multiple observed
measurements of systematically absent reflections should agree within their error
estimates and (ideally) have a mean = 0.0, thus it seems valid to include them in
measures of internal consistency such as Rmerge, Rmeas (strictly they should be
compared to 0 rather than their mean<Ih>), i suppose)
On the other hand, in looking at intensity statistics (as in truncate) the observed
intensities should be compared with their expectation values which will depend on
the space group (but still included perhaps). However there are so few of these
reflections that it won't make much difference (though it is more important to
separate centric& acentric reflections, as there may be quite a lot of
centrics)
ie "situation normal" (SNAFU?) - it is not going to make any difference
Phil
On 24 Jan 2011, at 09:13, Graeme Winter wrote:
Dear ccp4bb,
I had an interesting question from a xia2 user last week for which I
did not have a good answer. Here's the situation:
- spacegroup is P212121, which was specified on the command-line
- xia2 processes this as oP, assigns the spacegroup as P212121 before
running scala -> generates merging stats
- truncate removes systematically absent reflections
The end result is that there are fewer reflections in the output MTZ
file and hence used for refinement than are reported in the merging
statistics. The question is - what is correct? Clearly the effect on
the merging statistics will be modest or trivial as there were only ~
70 absent reflections, however removing them before scaling& merging
will also be a little fiddly.
I see three (or maybe four) options -
- truncate leave absent reflections in
- remove the absent reflections before scaling
- ignore this as "situation normal" (which is what xia2 currently does)
- (less helpful *) refine against intensities which includes the
absent reflections but matches the merging statistics
If this were my project I would probably opt for #3 but I can
appreciate that this is a question for the wider audience.
What do others think?
Many thanks in advance,
Graeme
* I mark this as less helpful because this is the wrong *reason* to
merge against intensities. There are clearly good reasons for this.
