Clemens,
I don't think anyone suggests to use unmerged data and such. I think
there is important difference in what Mohinder describes.
Let's say I have a ligand on symmetry axes and so it appears in two
conformations. If I reduce symmetry, there are two possible scenarios.
a. In lower symmetry, ligand still appears in two conformations. Shall
use higher symmetry.
b. In lower symmetry, ligand appears to be in single conformation (this
is what Mohinder says, if I am not mistaken). In this case, the true
symmetry is lower, and it is simply overwhelmed by the fact that most of
the structure (but not all) obeys higher symmetry.
I recall Bruce Foxman describing a b) case (I am sure there is more than
one example) for a small molecule crystal, where a single heavy atom had
higher symmetry than the rest of the molecule.
While in P21 reflections related by what will be now NCS operator are
almost identical, the small differences result in the ligand density
that is ordered. In P21212, these differences are averaged and the
information about disorder status of the ligand is lost. Afaic, it's
not really critical information, since it probably does not change the
conclusions regarding protein-ligand interactions.
Cheers,
Ed.
On Thu, 2010-10-21 at 16:57 +0100, Clemens Vonrhein wrote:
> Hi Herman,
>
> On Thu, Oct 21, 2010 at 05:31:51PM +0200, [log in to unmask] wrote:
> > If you process your data in a lower symmetry space group, you will have
> > more unique reflections, since reflections which are related by the
> > higher symmetry will be avaraged during scaling in a higher symmetry
> > space group (i.e. a 2fold or 3fold axis), while in lower symmetry space
> > groups they will not. So the observation to parameter ratio stays the
> > same and is only depending on resolution and solvent content.
>
> True - if you count Miller indices as observations. But if you think
> about information content than probably not (as you discuss below).
>
> > The question one has to ask of course is: are these reflections really
> > different, or are they the same only not averaged?
>
> Yes - by merging we're getting better data (better error estimate on
> the intensity due to higher multiplicity). So there isn't really
> independent information in 50% of the reflections if e.g. going from
> P21 to P1 - we've only increased the noise because the multiplicity of
> each reflection has been reduced.
>
> > In the latter case, you have more reflections, but not more
> > information. As Ed mentions, using tight "NCS" restraints would in
> > this case mimick the crystallographic symmetry.
>
> Apart from the (good) NCS argument, one could go even further:
>
> We could also just collect 36000 degree of data on a 7A Lysozyme
> crystal and refine against completely unmerged data. After all, why
> should we stop at removing only the some symmetry operators from our
> data merging ... lets get rid of all of them including th x,y,z
> operator and use unmerged data. Then we could refine Lysozyme with
> anisotropic hydrogens and no restraints against 7A data since we have
> a huge number of 'observations' ... right?
>
> But seriously: there is a difference in having reflections (H, K, L)
> and independent data (I, SIGI). Maybe we should talk more about
> (independent observations)/parameters ratio in the same way we
> look at depdencies of parameters (e.g. restraints on Bfactors etc).
>
> Cheers
>
> Clemens
>
--
Edwin Pozharski, PhD, Assistant Professor
University of Maryland, Baltimore
----------------------------------------------
When the Way is forgotten duty and justice appear;
Then knowledge and wisdom are born along with hypocrisy.
When harmonious relationships dissolve then respect and devotion arise;
When a nation falls to chaos then loyalty and patriotism are born.
------------------------------ / Lao Tse /
|