AUTOMATIC IDENTIFICATION OF MOLECULAR SIMILARITY USING REDUCED-GRAPH REPRESENTATION OF CHEMICAL-STRUCTURE

This paper describes an approach for automatic identification of the similar structural features among molecules. Here, each structure is described with an abstracted chemical graph of which each node describes a functional atomic group and the edge is weighted by the topological path length (the number of bonds) between two functional atomic groups to be considered. Isosterism for functional groups or substructures to be considered can be defined as a kind of knowledge file, the contents of which depend on the problem. A clique-finding algorithm was used for the search of common or similar structural features. Details of the algorithms will be discussed here, and a couple of illustrative examples in structure-activity problems will be shown for the applications.