[phenixbb] RFZ is higher than TFZ

[Nathaniel Echols]

On Thu, Jul 12, 2012 at 12:38 AM, 周文昌 <wenchangyu2006 at gmail.com> wrote:
> I had one problem using phaser-MR to solve my structure. Phaser-MR gives me
> one solution with TFZ=15.4 and LLG=165 which seems to be promising but
> phenix.refine does not give me a low R values (0.4884/0.5424). I went back
> and checked the solution from phaser-MR, I found RFZ=18.8, higher than TFZ.
> Does anyone think this is a problem to solve my structure?
>
> The space group is C121, I believe there are 4 copies of my search model in
> ASU.

What is the resolution, and what is the sequence identity of the search model?

I don't know whether the high RFZ means anything, but the first thing
I would check is whether the packing of the MR solution makes sense;
Phaser will do a good job preventing molecules from overlapping
(unless you changed it to use much less conservative criteria), but it
won't guarantee that the molecules pack into a nice continuous
lattice.  If there are big gaps, it may not be a real solution.

I think we need to add an entire FAQ page for high R-factors.
However, there are an increasing number of tools and/or tricks in
Phenix to deal with problems like this - in approximate order of
runtime:

- run phenix.refine for more cycles - at least 20.  See this paper for
an example of why this can be important:
http://journals.iucr.org/d/issues/2012/07/00/kw5044/index.html
- use phenix.morph_model (refer to previously mentioned paper)
- use phenix.den_refine (or DEN in CNS)
- use MR-Rosetta

Also, if you're willing to share your data with us, we're collecting
difficult-to-refine structures for benchmarking new methods.  (It will
of course be kept confidential.)

-Nat