[MMTK] point group symmetry

Konrad Hinsen hinsen@cnrs-orleans.fr
06 Jun 2002 16:34:48 +0200


Margot Ernst <margot.ernst@univie.ac.at> writes:

> modes). With low coarseness and high deformation threshold, using up
> to 4 modes, I get nearly symmetrical domains, but as I increase c (
> 10 and larger) the symmetry gets lost dramatically. So, after my
> first mail I was in fact able to find reasonably symmetrical
> domains, but I fail to understand why the symmetry gets lost totally
> at higher c. This far, including up to 4 modes and fine-tuning the c

The clustering algorithm doesn't use symmetry, nor even the positions
of the cubic regions. The only information it uses is the rigid-motion
parameters of the cubes (6 per mode). It starts with the two most
similar cubes and then adds more and more until no cubes are left that
are within the coarseness limit. Whatever is left after the first
cluster/domain has been assigned is the starting point for locating
the remaining clusters/domains.

At high coarseness levels, the algorithm will assign too many cubes to
each domain, starting with the first, which becomes the largest one
(in fact at some coarseness level the whole protein will become a
single domain). If, for example, you have two symmetric domains, then
by increasing the coarseness level the first one will grow and the
second one shrink, simply because the first one has the first chance
to grab cubic regions.

The domain analysis process is not meant to be fully automatic, and
not every coarseness level gives a meaningful result. In fact,
symmetry, contiguousness of the domains, and other such properties can
and should be used to decide which proposed domain decomposition is
reasonable.

It is also worth watching the diagram at the bottom of the domain
analysis window. It shows one line for each cubic region, and the
values on the axes are the rigid-body motion parameters. Regions that
form one domain should be visually "close" and form a recognizable
group. If they don't, choose different parameters.

Konrad.
-- 
-------------------------------------------------------------------------------
Konrad Hinsen                            | E-Mail: hinsen@cnrs-orleans.fr
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.56.24
Rue Charles Sadron                       | Fax:  +33-2.38.63.15.17
45071 Orleans Cedex 2                    | Deutsch/Esperanto/English/
France                                   | Nederlands/Francais
-------------------------------------------------------------------------------