A graph theory-based “expert system” methodology for structure-activity studies

A new graph theory-based methodology is proposed for the statistical evaluation of the link between the structure and the chemical/biological activity of organic molecules. The computer-aided analysis involves the heuristic processing of molecules as chemical graphs which are decomposed into their component subgraphs. Common topological features among the subgraphs are then statistically isolated, and a set of rules is developed that can be used to explain the activities of the analyzed compounds as well as predict the activities of new compounds. The validity of the methodology is demonstrated by its application to actual experimental data.