Hi Ed, I've actually run that exact test in phenix as an exercise to prove to my PI the validity of occupancy refinement. Though as a disclaimer it was a 1.2 angstrom data set and this was an alternate conformation situation. I ran different input occupancies without occupancy refinement and measure the difference density peak values and average B-factors and ended up with the same occupancy ratio that the program's occupancy refinement spit out. Of course this might not hold true if someone is refining the occupancy of a ligand that is partially bound without an alternate option (i.e. total occupancy <1). I haven't tested that one systematically yet though I suspect Pavel has probably already done this at some point.
Cheers, Katherine On Mon, Jun 4, 2012 at 7:35 AM, Ed Pozharski <epozh...@umaryland.edu> wrote: > > Is it reasonable to refine occupancy in phenix at 2.2 A resolution? > > Implementations may differ, but imgo refining occupancy at 2.2A > resolution is not very reasonable under most circumstances, as it will > correlate strongly with the B-factor. A reasonable approach might be to > fix occupancy at different levels and get a series of refined models. > Then you look at (i) B-factor behavior and see at what occupancy it > matches the surrounding atoms and (ii) difference density (my > unsubstantiated theory is that if you plot it against occupancy it > should have a central flat region where B-factors are capable of > compensating and two linear regions on extreme ends which should allow > to extrapolate the true value. > > Refmac does occupancy refinement. It's quite fast, so you may try > randomizing the initial value and get some idea about convergence. > > Cheers, > > Ed. > > > > -- > Oh, suddenly throwing a giraffe into a volcano to make water is crazy? > Julian, King of Lemurs >
