b)how
does PHENIX/refine handle the fobs = 0 reflections
in either case?
phenix.refine outputs two series of maps:
- one pair of (2mFo-DFc, mFo-DFc) maps is computed using original
untouched Fobs;
- and the other pair of (2mFo-DFc, mFo-DFc) maps is computed using
manipulated Fobs, where missing Fobs are "filled" in with DFc (only for
map calculation!!!). To avoid any confusion, this is clearly indicated
in output MTZ file with map coefficients.
As I said, by default missing Fobs are filled with DFc, but there are
other options available to play with, such as <Fobs>, simply Fc,
random numbers, and... I forgot the whole list but I re-call I
implemented a bunch of possibilities.
I did a number of tests (and even documented it!) and actually spent
quite a bit of time playing with this... My observation was that
filling with DFc or <Fobs>, or with even random numbers
(generated around <Fobs>) did NOT produce any visible difference
between each other (indicating the overwhelming importance of phases) -
all of them were almost equally good.
Sure, there were the cases where "filled" maps were giving much more
interpretable maps NOT only because of bias, but just because of
eliminating data incompleteness effects by putting "at least something"
into missing Fobs slots. I've seen bias as well.
Overall feeling is: it is the best to look at BOTH maps: filled and not
filled, to help overcoming a difficult cases and still staying on a
safe side.
You can play with this in phenix.refine if you like.
Overall, this whole thing requires more tests, and care before any
generalizations is made.