Research article

Evaluation of a Bayesian inference network for ligand-based virtual screening

Beining Chen , Christoph Mueller and Peter Willett

Krebs Institute for Biomolecular Research, Departments of Chemistry and of Information Studies, University of Sheffield, Sheffield, S10 2TN, UK

author email corresponding author email

Journal of Cheminformatics 2009, 1:5doi:10.1186/1758-2946-1-5

Published:	29 April 2009

Abstract

Background

Bayesian inference networks enable the computation of the probability that an event will occur. They have been used previously to rank textual documents in order of decreasing relevance to a user-defined query. Here, we modify the approach to enable a Bayesian inference network to be used for chemical similarity searching, where a database is ranked in order of decreasing probability of bioactivity.

Results

Bayesian inference networks were implemented using two different types of network and four different types of belief function. Experiments with the MDDR and WOMBAT databases show that a Bayesian inference network can be used to provide effective ligand-based screening, especially when the active molecules being sought have a high degree of structural homogeneity; in such cases, the network substantially out-performs a conventional, Tanimoto-based similarity searching system. However, the effectiveness of the network is much less when structurally heterogeneous sets of actives are being sought.

Conclusion

A Bayesian inference network provides an interesting alternative to existing tools for ligand-based virtual screening.