Multidimensional scaling of combinatorial libraries without explicit enumeration

A novel approach for the multidimensional scaling of large combinatorial li braries is presented. The method employs a multilayer perceptron, which is trained to predict the coordinates of the products on the nonlinear map fro m pertinent features of their respective building blocks. TI-Lis method lim its the expensive enumeration and descriptor generation to only a small fra ction of products and, in addition, relieves the enormous computational eff ort required for the low-dimensional embedding by conventional iterative mu ltidimensional scaling algorithms. In effect, the method provides an explic it mapping function from reagents to products, and allows the vast majority of compounds to be projected without constructing their connection tables. The advantages of this approach are demonstrated using two combinatorial l ibraries based on the reductive amination and Ugi reactions, and three desc riptor sets that are commonly used in similarity searching, diversity profi ling and structure-activity correlation. (C) 2001 John Wiley & Sons, Inc.