A similarity-based data-fusion approach to the visual characterization and comparison of compound databases

A low-dimensional method, based on the use of multiple fusion-based similarity measures, is described for graphically depicting and characterizing relationships among molecules in compound databases. The measures are used to construct multi-fusion similarity maps that characterize the relationship of a set of 'test' molecules to a set of 'reference' molecules. The reference set is very general and can be made of molecules from, for example, the set of test molecules itself (the self-referencing case), from a small library or large compound collection, or from actives in a given assay or group of assays. The test set is any collection of compounds to be analyzed with respect to the specified reference set. Multiple fusion similarity measures tend to provide more information than single fusion-based measures, including information on the nature of the chemical-space neighborhoods surrounding reference-set molecules. A general discussion is presented on how to interpret multi-fusion similarity maps, and several examples are given that illustrate how these maps can be used to compare compound libraries or collections, to select compounds for screening or acquisition, and to identify new active molecules using ligand-based virtual screening.