Abstract: MATH/CHEM/COMP 2002, Dubrovnik,
June 2429, 2002

MOLGEN
4.1 and its
Application in Combinatorial Chemistry
Joachim Braun,
Ralf Gugisch, Adalbert Kerber, Reinhard Laue,
Markus Meringer, and Christoph Rücker
Department of Lehrstuhl Mathematik II, Universität
Bayreuth,D95440 Bayreuth, Germany Combinatorial
chemistry experiments can clearly be optimized through the use of
mathematical simulations. Therefore we provided the new version of MOLGEN
with several tools for the application in combinatorial chemistry. A
new kind of structure generator has been developed, that allows to construct
virtual combinatorial libraries. The input of this structure generator called
MOLGENCOMB is a mathematical description of reactions and reactants. Using a
canonical form our software is able to compare combinatorial libraries, in particular
for testing whether a given real library is a subset of a constructed virtual
library. In
order to predict physical, chemical or biological properties for the virtual
libraries, we implemented various molecular descriptors that serve as input
for regression analysis. At the moment there are 252 arithmetical,
topological and geometrical descriptors included in our software. Regression
analysis correlates molecular descriptors with measured properties of the
real library. Regression methods are provided by the statistics package
"R" which is accessed directly from MOLGEN's graphical user
interface. So far multilinear regression and regression trees are available
in order to suggest promising candidate structures for the target property.
Financial support by the BMBF and the DFG is
gratefully acknowledged 