Inference networks for molecular database similarity searching

Abstract

Molecular similarity searching is a process to find chemical compounds that are similar to a target compound. The concept of molecular similarity play an important role in modern computer aided drug design methods, and has been successfully applied in the optimization of lead series. It is used for chemical database searching and design of combinatorial libraries. In this paper, we explore the possibility and effectiveness of using Inference Bayesian network for similarity searching. The topology of the network represents the dependence relationships between molecular descriptors and molecules as well as the quantitative knowledge of probabilities encoding the strength of these relationships, mined from our compound collection. The retrieve of an active compound to a given target structure is obtained by means of an inference process through a network of dependences. The new approach is tested by its ability to retrieve seven sets of active molecules seeded in the MDDR. Our empirical results suggest that similarity method based on Bayesian networks provide a promising and encouraging alternative to existing similarity searching methods.