Mining Fragments with Fuzzy Chains in Molecular Databases

This paper discusses methods to discover frequent, discriminative connected subgraphs (fragments) in a database of molecular structures. We present an extension to a well-known algorithm that allows for the discovery of fragments that contain chains of atoms of varying length. This is particularly important for real-world applications (for example drug discovery or synthetic success prediction) where the exact length of chains connecting two or more otherwise rigid substructures is not critical for the biological or chemical activity of the overall substructure. We demonstrate how the proposed extension successfully discovers fragments with several polymethylene bridges.