Partition-based selection

This article concentrates on diversity-related methods in which the analysis involves a partitioning into discrete cells of compounds or data derived from compounds. A detailed discussion is given of two major approaches. In the first approach, each compound is assigned to only one cell, and two methods that use partitioning by molecular/atomic properties are described (DPD; BCUT/DiverseSolutions). In the second approach, each compound may exhibit multiple values of the property being partitioned, particularly when conformational flexibility is taken into account, and therefore may belong to many partitions; two methods that use potential three-dimensional pharmacophores as a molecular similarity/diversity measure are described (PDQ and ChemDiverse). The use of these methods for database analysis, subset selection and combinatorial library design is discussed.