On the 7th International Conference on Chemical Structures Jeroen Kazius has a poster on finding discriminative substructures, that is, molecular fragments which can be discriminate between two acitivity classes. The software is released as Gaston, is written in C++ and has the GPL license.

Later I encountered MoSS which has the same goal, but uses a different algorithm. MoSS is written in Java and uses the LGPL license. MoSS reads STN and SMILES as input, which might not be optimal for all users, so a CDK port comes to mind.

This post was originally posted as https://chem-bla-ics.blogspot.com/2005/11/open-source-data-mining-in.html.