[MMTK] Fitting data?

Jon Wright wright at esrf.fr
Tue Mar 23 16:46:17 CET 2004

Hi everyone,

I've just downloaded and had a quick look at the MMTK package, which 
looks very impressive (thanks!). I'm wondering if it would be possible 
to use it as a basis for fitting some powder diffraction data from 
crystalline protein samples... The idea is to add an energy term 
proportional to a chi^2 function which describes the fit to the data. I 
think I can (eventually) supply the value of this term, and it's 
derivatives with respect to x,y,z co-ordinates of the protein atoms in a 
unit cell (which could be transformed into an infinite universe). There 
is a slight hitch, that for crystallography the structure is described 
with the atoms smeared out by a "thermal factor", but given the poor 
quality of powder data this can be just left as a constant and one could 
forget the time average versus instantaneous configuration question. The 
manual (p13) mentions a developers guide, which I did not manage to find 
yet - would someone direct me?

 From browsing the source, it looks like it should be possible to get 
started with the example MMTK_harmonic_oscillator.c etc, with some bits 
and pieces to specify the powder dataset, unit cell, space group etc in 
the constructor in the associated .py file? I wanted to contact the list 
before diving in, things like unit cell and space group probably already 
have proper homes in other parts of the library, but the treatment of 
symmetry seems a bit vague so far. Currently we're using a package which 
is "state of the art" for fitting powder data, and has been recently 
modified to treat proteins. The results are great from the point of view 
of fitting the powder data, but make me nervous when I look at the 
resulting structures.

We would benefit greatly if we could do some cross validation, and would 
love to run some simulated annealing procedures where we try to find 
good stereochemical configurations which happen to fit our data as well. 
Ideally I need some advice from someone who knows more about molecular 
dynamics, because my background is mainly limited to powder diffraction. 
It would be very interesting to generate a complete unit cell, including 
water molecules, to address some questions about low resolution 
crystallographic data and "solvent scattering" contributions to low 
angle peaks.

Any help, advice or hints would be greatly appreciated!

Thanks in advance,


PS: If only you could understand how happy I was to find out that I can 
"fix" the stereochemistry of a model by running it through a minimiser 
with a little command line script! This is already a great help if I can 
cycle models back and forth between the powder software and a better 
stereochemical minimiser. Until today there was a lot of mouse clicking 
needed to accomplish the same thing...

Dr. Jon Wright
European Synchrotron Radiation Facility,
BP-220, F-38043, Grenoble, France

More information about the mmtk mailing list