On Fri, 22 Oct 2010 13:38:07 -0700
Ethan Merritt <[log in to unmask]> wrote:
>
> Unfortunately for the current discussion, I do not find any practical
> comparison to Babinet models in either paper. Both start with the
> implicit assumption that a flat, masked model is better and then
> proceed to explore how best to determine its parameters. The Fokine
> and Urzhumtsev paper does state that "the [Babinet] exponential
> scaling model is handicapped at higher resolution", but does not
> present any examples or comparisons to document what effect this has
> in practice.
>
> The same is true for the very recent paper by Fenn, Schneiders, and
> Brunger Acta Cryst. (2010). D66, 1024-1031
> [ doi:10.1107/S0907444910031045 ] This paper presents the theoretical
> basis for the Babinet treatment and for several recent hybrid mask
> treatments that get away from describing the solvent region as
> "flat". But the paper does not include a Babinet model in the tables
> of comparative results.
>
> Do you know of any published or unpublished results that compare
> the R factors achieved by Babinet treatment with those obtained from
> the state-of-the-art mask models?
>
> I'll cc this to Tim Fenn in case he wants to chime in with additional
> data that wasn't included in the recent paper.
>
Sorry for my late reply, and I'll try to explain a little more of our
thinking in the paper you brought up. We don't have data to the effect
you're asking about, but I feel the implementation outlined in the
paper is the combination of the flat model *and* Babinet's principle.
When we were initially thinking about how to go about things, we drew
out a physical picture - which we liked so much we included in the
paper as Figure 2. This made it clear to us that the phase of Fm -
which is just one minus the realspace bulk solvent mask as it is
usually generated, should be inverted, *not* Fc. And the definition of
Babinet's, I think, backs this up: the diffraction of the bulk solvent
that we want to add in is equivalent to adding the inverse of its
complement, which is the inverse of the protein region as if it
scattered as bulk solvent (NOT as protein).
Some history that also explains how we got to this: when Mike and I
first tinkered with bulk solvent methodology, we had an admittedly slow
implementation - we would expand the model to P1, then generate the
solvent mask. This got us to our primary motivation with the new mask
definitions: they were differentiable, but we wanted to speed things up
while avoiding the need to introduce asymmetric unit definitions in our
code (yes, I'm that lazy) so we could apply symmetry in reciprocal
space, which is must faster. So we started thinking of inverting the
mask, or just using a protein mask (then we wouldn't care where the
boundary of the asymmetric unit is - quiz question: why?), which
naturally led us to discussions of Babinet's and the realization that
it wasn't that crazy of an idea. So I think the implementation has a
few advantages, since it combines the Babinet principle with the flat
model (and the new models, which have some nice bonuses with respect to
atomic gradients), makes coding easy and it runs fast (I have some
timings to illustrate the latter, I'd be glad to show them to anyone
thats interested).
Sorry for the long-windedness, but I hope this answers the question:
Babinet's vs. the flat model? Use them together! ;)
Regards,
Tim
--
---------------------------------------------------------
Tim Fenn
[log in to unmask]
Stanford University, School of Medicine
James H. Clark Center
318 Campus Drive, Room E300
Stanford, CA 94305-5432
Phone: (650) 736-1714
FAX: (650) 736-1961
---------------------------------------------------------
|