Definitions of "dissimilarity" for dissimilarity-based compound selection

Dissimilarity-based compound selection involves identifying a database subset in which the constituent compounds are as dissimilar to each other as possible, thus ensuring coverage of the full range of structural diversity in the original database. This paper provides a quantitative comparison of four different definitions of dissimilarity. Experiments with three different measures of diversity demonstrate that the effectiveness of the selected subset is affected by the definition of dissimilarity that is used, but that it is not possible to identify one such definition as being consistently superior to any other.