Cooperation with ALTANA Pharma AG

Computational Chemistry-Library JOELib JOELib is a platform independent open source computational chemistry package written in Java. This package uses graph based structures to represent and modify molecular structures and allows not only to calculate a huge number of descriptors, but also 'SMiles ARbitrary Target Specification' (SMARTS) substructure search. Within this project also a number of algorithms for the maximum common substructure search on molecular graphs have been developed and investigated. The JOELib sources and the binary distribution can be downloaded from Sourceforge. There is also a Brief Tutorial and an API Documentation available.

Optimization of Combinatorial Libraries The optimization of combinatorial libraries is typically a complicated, often even NP-hard, and sometimes also a multi-objective optimization problem. In this cooperation we address the combinatorial library design problem by means of multi-objective evolutionary algorithms, which allow us to deal with a number of objectives at the same time, resulting in a whole set of so called Pareto optimal libraries instead of just a single library as solution. Possible objectives for the optimization of combinatorial libraries may include the diversity of the library, the activity, the similarity to given leads, the overlap to existing libraries and/or the similarity to a target distribution for certain properties. The possible EA representation to be optimized varies depending on the internal library structure used. In case of given scaffolds and RGroups the EA representation may be based on sparse subsets, full arrays or plates, each representation having certain characteristics regarding the complexity of the resulting optimization problem. Currently, we are examining a number of multi-objective optimization techniques, not only multi-objective EAs, but also weight aggregation, goal programming and interactive techniques like the Tchebycheff, the Step or the reference point method. Depending on the multi-objective optimization technique and the representation used we are experimenting with genetic algorithms, evolution strategies, population based incremental learning, particle swarm optimization and many more. Further details on the optimization algorithms used can be found on the JavaEvA webpage. The left hand side figure shows an exemplary Pareto optimal solution of a sparse array (red dots) trying to maximize the diversity in a two dimensional real-valued descriptor space while at the same time minimizing the distance to three given leads (green dots). In the upper part the distribution of the resulting library regarding a third descriptor is given.

Contact

Felix Streichert, Tel.: (07071) 29-70436, streiche at informatik.uni-tuebingen.de