Inference networks for chemical similarity searching

Abstract

Similarity searching is becoming the simplest tool available for similarity-based virtual screening of chemical databases. Over the years many methods have been developed. A variety of similarity metrics have been introduced, but by far the most prominent is the Tanimoto coefficient. Currently, Bayesian classifiers are increasingly widely used for virtual screening of chemical databases. In this paper, a novel similarity searching approach using inference Bayesian network is discussed. The retrieved of an active compound is obtained by means of an inference process through a network of dependences. Experiments on MDDR demonstrate that similarity approach based on Bayesian inference networks outperforms the similarity search approach with Tanimoto coefficient and offer promising alternative to existing similarity search approaches.