ABSTRACT
Ligand-protein binding interactions are primarily governed by three-dimensional (3D) spatial interactions. Consequently, in an ideal world one would wish to apply 3D descriptors when undertaking molecular selection for most computeraided molecular design (CAMD) problems. Unfortunately, even the fastest approaches typically require molecular superimposition in coordinate space before most 3D molecular descriptors can be utilized (I). As a result, the CPU time needed for such calculations is too large to cope with the virtual data set sizes of most combinatorial library calculations or compound selection exercises. To overcome this problem, a number of descriptors with 3D content have been devised that circumvent the need for molecular superposition. In their original guises such descriptors were primarily designed as shape similarity measures based on matching interatomic distance distributions (2-5). It was soon realized, however, that
elements of molecular chemistry could be incorporated to produce full pharmacophoric fingerprints (6-10). A pharmacophore is defined here as a critical threedimensional geometric arrangement of molecular fragments fonning a necessary but not sufficient condition for biological activity ( 11, 12). The use of such descriptors has formed a mainstay of ligand-based virtual screening for much of the last decade. These descriptors have proven to be excellent for divorcing the 3D structural requirements for biological activity from the 2D chemical makeup of a ligand. The resulting measures are thus able to exploit even limited data regarding a target to discover structurally novel active chemotypes. This proven ability, together with calculation speeds that pennit their use on data sets deemed too large for most other 3D measures, make them attractive as combinatorial library descriptors. In this chapter we further highlight reasons for their utility and detail the techniques applied thus far in their exploitation.
