Abstract. Identifying the structure of unknown molecules is an important activity in the pharmaceutical industry where it underpins the production of new drugs and the analysis of complex biological samples. We present here a new method for automatically identifying the structure of an unknown molecule from its nuclear magnetic resonance (NMR) spectrum. In the technique, an ant colony optimization algorithm is used to search iteratively the highly-constrained space of feasible molecular structures, evaluating each one by reference to NMR information on known molecules stored (in a raw form) in a database. Unlike existing structure elucidation systems, ours: does not need prior training or use spectrum prediction; does not rely on expert rules; and avoids enumeration of all possible candidate structures. We describe the important elements of the system here and include results on a preliminary test set of molecules. Whilst the results are currently too limited to allow parameter studies...
Caroline Farrelly, Douglas B. Kell, Joshua D. Know