Stochastic algorithms for maximizing molecular diversity

A common problem in the emerging field of combinatorial drug design is the selection of an appropriate subset of compounds for chemical synthesis and biological evaluation. In this paper, we introduce a new family of selection algorithms that combine a stochastic search engine with a user-defined objective function that encodes any desirable selection criterion. The method is applied to the problem of maximizing molecular diversity, and the results are visualized using Sammon's nonlinear mapping algorithm. By separating the search method from the performance metric, the method can be easily extended to perform complex multiobjective selections in advanced decision-support systems.