[MMTK] MMTK: Ready for Prime Time?

Konrad Hinsen hinsen@cnrs-orleans.fr
Fri, 25 Aug 2000 19:26:34 +0200


> I'm interested in finding out if (and how many) people are using MMTK
> for "production" molecular modeling work, and what their experience has

At least me! I have been using MMTK for all my research for the last
four years, and I plan to continue.

> been.  In playing with the program during the last few days I have been
> very impressed with existing capabilities (such as the Amber force-field
> and normal mode calculations) but there are still some big holes.  Also,

Certainly. Until now, the majority of the code has been written by
myself for my own needs, so the MMTK feature list reflects the topics
that I have worked on. There have of course been contributions from
others, for example the molecular surface module was contributed by
Lutz Ehrlich. But many more contributors would be necessary to make
MMTK a complete simulation and modelling environment.

> the conjugate gradient minimizer seems very slow to converge ( in
> execution time - not necessarily steps) compared to other mimizers I've
> used.

It is indeed slow, because I never optimized it. Neither the
implementation for one CG step (essentially one line minimization) nor
the "end game" strategy for converging to the minimum is optimal. I
have used it exactly four times until now for my research, so I
couldn't justify the effort for optimization. And I am by no means a
minimization specialist.

> Are there any efforts underway to implement the following features into
> MMTK?
> 
> - An implicit solvation potential energy function

That would be a big project, already for deciding which one to
implement. This definitely requires someone familiar with the
techniques.

> - Amber force field entries (suitable atom types + cross terms) for
> common functional groups found in proteins ligands (chloride,
> sulfonamide, various heteroatom rings, etc.)

If it exists for Amber it can be used with MMTK - MMTK reads the
standard Amber parameter files. Krzysztof Murzyn from Cracow, who has
spent six months next to me working on an analysis program based on
MMTK, has also written a converter that generates MMTK database
entries from Amber topology files, and he has also worked on OPLS
support (i.e. database entries). So going from Amber to MMTK should be
pretty easy soon, no matter what the system is.

What do you mean by "cross terms"? The Amber force fields doesn't have
any cross terms in the standard sense (i.e. coupling between bonds and
angles etc.). Neither does MMTK.

> - Automated type and charge assignment (I've done some work on the
> former using Python)

There is a charge fitting module in MMTK which takes electrostatic
potential surface data, but you need some quantum chemistry program to
provide this data, so this is far from automated. As for type
assignment, I am not even familiar with the techniques used for that.

> - Additional optimizations (such as not calculating forces between fixed
> atoms).

That could be done rather easily in the sloppy way (leaving out the
energy terms between fixed atoms), and with some more effort also more
properly (calculating the energy terms once in order to get the
correct total energy).

> - A faster CGM.

Shouldn't be difficult for someone who is familiar with the methods.

> I realize that some of these are tough problems, but the hope of a
> freely available Python based molecular modeling system is a dream that
> I would love to help make a reality.

Me too! The biggest problem is manpower. Implementing some non-trivial
technique in MMTK requires someone who

- knows the technique and the science behind it
- is a sufficiently competent programmer
- has sufficient Python experience

Unfortunately there are not so many people who meet these
requirements, and it doesn't help that implementation work is not yet
recognized as a scientific activity. The well-known simulation
packages have all been written by large and well established groups
over many years. Much of the actual implementation work has been done
by PhD students supervised by more experienced people in the same
group. MMTK, until now, has essentially been written by one person
over four years on postdoc contracts. I do have a permanent position
now, but I am in a new group that is still small and has no PhD
students (quite a common situation in France). I do expect other
people in this group to contribute to MMTK in the future, but the
available manpower as well as the diversity of research topics will be
limited in the forseeable future.

Significant progress thus requires other groups to join the
development efforts. I hope that this will happen, and I will happily
provide support to those who want to implement something in MMTK.

If anyone has ideas for increasing the developer base, I'd like to
hear them. Would it help to move development to a more visible place
(SourceForge, bioinformatics.org)? Are there better ways to spread
information about the project? How could scientists be convinced
of the value of the OpenSource approach?

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
-------------------------------------------------------------------------------