Cell-Based Methods for Sampling in High-Dimensional Spaces

Abstract

We regard the compounds in a combinatorial library of potential drugs as points in a Euclidean space of many dimensions, where the coordinates of a point represent properties used to characterize the compound it represents. By decomposing the space into certain advantageously selected Voronoi cells (not by the standard decomposition into hypercubes), and by choosing within each cell (e.g., for closer scrutiny, synthesis, etc.) a point near its center, one obtains conveniently small subsets that are highly representative of the larger library.