SIMILARITY SEARCHING ON CAS REGISTRY SUBSTANCES .2. 2D STRUCTURAL SIMILARITY

Chemical Abstracts Service (CAS) is exploring approaches for similarity (''fuzzy-match'') searching on CAS Registry substances. Experimental software is being developed to identify, analyze, and perform similarity searches on various characteristics of an integrated set of 2D, 3D, and molecular property data for samples of Registry substances. Earlier results have indicated that searching on global molecular property features such as ionization potentials and van der Waals' volumes appears to detect ''chemical'' (isosteric) similarity and that searching on generic atom triangle geometric features provides a significant amount of shape and size similarity. More recently, we have been exploring possibilities for 2D global and local structural similarity on Science and Technology Network (STN) structure files. One possible approach involves one or more fragment-based searches using the existing STN 2D substructure screen fragments, optionally followed by existing connectivity-based (atom-by-atom) search on generic structure representations of candidates obtained in the fragment-based screening step. Fragment-based searches using various screen classes such as augmented atoms and bond sequences provide for different views of 2D) structural similarity. Connectivity;based searching of generic 2D structures allows for a considerable amount of flexibility in a user's definition of similarity. This paper will discuss recent results of a comparison of the effectiveness of the various STN screen classes in fragment-based similarity searching using the Tanimoto coefficient and will illustrate the STN capabilities for connectivity-based similarity searching on an answer set.