[phenixbb] Refining anomalous groups

[Nathaniel Echols]

On Sat, Jan 12, 2013 at 10:13 AM, Pavel Afonine <pafonine at lbl.gov> wrote:
> My understanding is that this option will select all atoms that are heavier
> than some defined value, and then it will refine f' and f'' for them. This
> may mean that you will be refining f' and f'' for atoms that were not
> anomalously scattering; that is you will be refining parameters that do not
> need to be refined. While this may be harmless (though someone needs to
> prove it) I believe it is suboptimal: 1) runtime wise, and 2) unnecessary
> refined parameters may mop up other model errors. Analogy could be: refining
> occupancy of fully occupied atoms.

Technically every atom has anomalous scattering - it's simply
unmeasurable for most of them at the wavelengths we use.  But as Felix
points out, S atoms are frequently visible in the anomalous difference
map, and when this isn't the case, it's usually because much stronger
anomalous scatterers dominate.  Therefore the anomalous difference map
by itself isn't a reliable means to identify the anomalous atoms.  We
could do it iteratively using the LLG map from Phaser, which I've
tried and works quite well - but at a huge cost in runtime.  Simply
picking the heavier atoms and refining those instead was the simplest
and fastest solution.  The runtime does not appear to be excessive.

The main problem is indeed with the soaking up of occupancy
differences - if the real scattering is not modeled correctly, f'
typically refines to unrealistic values to compensate.  I can think of
a couple of possibilities for coping with this, but the best approach
would be to also refine the occupancies of any atoms which will be
anomalous (perhaps with the exception of those which are part of
macromolecules), since that will be done before the anomalous
refinement.  (Although I'm not sure whether the occupancy refinement
has a reciprocal effect on f'.)

-Nat