[Cdk-devel] Re: [Jchempaint-devel] Force Field Implementation in CDK - Please advise

[Peter Murray-R. <pm...@ca...>]

At 12:58 23/10/2003 +0200, Christoph Steinbeck wrote:
>Peter Murray-Rust wrote:
>
>>I also think it's very important. There are several reasons:
>>- I could use an off-the-shelf forcefield for building molecules. Nothing=
=20
>>special, just to relieve the worst strain in an initial configuration
>
>Right - it will certainly be an enrichment of CDK's functionality. I'm=20
>also thinking about making it flexible enough to work with 2D and 3D=20
>equations and parameter sets in order to supplement SDG.

At present we have proof-of-concept for building up molecules from a=20
fragment library. This will often produce locally bad geometry or distant=20
interactions. So a quick minimise to get rid of the worst would be very=20
valuable.

>>- We are currently discussing the abstraction of forcefields at=20
>>fsatom.org. I suspect this is too large a topic, but I am interested in=20
>>the idea of reading in a functional form and generating code to support=20
>>that. Not everyone agrees
>
>I was not aware of this site and project. Sounds interesting. Subscribed.
>The idea that you sketch sounds great but would certainly be at least one=
=20
>full Ph.D. thesis, I suppose.
>
>>Can we assume that if the atomTypes are known then the construction of=20
>>the FF is algorithmic? i.e. something like:
>>- assign atomTypes
>>- iterate through structure identifying:
>>         bonds
>>         angles
>>         dihedrals
>>         oop
>>         non-bonded contacts
>>         electrostatics
>>- assign numeric parameters to each of the above based on:
>>         - a lookup table of allowed atomType combinations
>>         - or values calculated by interpolation
>>This should all be automatic.
>
>Exactly like this.

I have used MM2 a lot. (some years ago). The problem was then when you come=
=20
to almost any torsion (e.g. C.ar-C.3-S-C.2) it would not be in the lookup=20
table. So we need a universal force field. I am out of touch here and=20
suggest you could ask on CCL or elsewhere.

>>The you need to decide:
>>- do you want derivatives?
>>- do you want second derivs?
>
>I'm not an expert here but from some basic reading I'm getting the=20
>impression that, for the really fast and simple minimization routines like=
=20
>steepest decent one needs first order derivatives. Simulated Annealing=20
>does not need them, but is certainly slower and more difficult to=
 parameterize.

I agree that first derivatives are important. They should be liftable from=
=20
many source codes by now.

>>I think if this is a CS student they should concentrate on the numeric=20
>>stuff and not  so much time on crafting the forcefield to fit reality.
>
>We do not intend to do the latter. We need something basic which works and=
=20
>which potentially is a starting point for later optimizations. I guess the=
=20
>20 weeks might just be enough to get a very basic but functional version=20
>of our engine.
>
>>So this would be very useful and hopefully could be called with a single=
=20
>>method (customisable if necessary).

I would ask on the lists but if that fails I would probably have a look at=
=20
Tinker, even though it's F77. It's cleanly written. MM2 when I last looked=
=20
(many years ago) had

>Absolutely. Something like
>
>interface ForceFieldEngine
>{
>         public setMolecule(Molecule m);
>         public Molecule getMolecule();
>         public minimize();
>// or maybe just?
>         public Molecule minimize(Molecule m)
>         public setParameters (ForceFieldParameters ffp);
>
>}
>
>class DreidingForceFieldEngine implements ForceFieldEngine
>{
>}
>
>Cheers,
>
>Chris
>
>--
>Dr. Christoph Steinbeck (e-mail: c.s...@un...)
>Groupleader Junior Research Group for Applied Bioinformatics
>Cologne University BioInformatics Center (http://www.cubic.uni-koeln.de)
>Z=FClpicher Str. 47, 50674 Cologne
>Tel: +49(0)221-470-7426   Fax: +49 (0) 221-470-7786
>
>What is man but that lofty spirit - that sense of enterprise.
>... Kirk, "I, Mudd," stardate 4513.3..