Ligand-based virtual screening using Bayesian networks.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Similarity-based virtual screening is crucial for identifying potential drug candidates.
• Bayesian inference networks (BINs) are effective for structurally homogeneous active molecules.
• BINs show limitations with structurally heterogeneous datasets.

Purpose of the Study:

• To introduce a novel Bayesian belief network (BBN) model.
• To address the limitations of BINs in handling structurally diverse active molecules.
• To evaluate the performance of BBNs in virtual screening.

Main Methods:

• Development of a Bayesian belief network (BBN) model.
• Simulated virtual screening experiments using MDDR, WOMBAT, and MUV datasets.
• Comparison of BBN performance against Bayesian inference networks (BIN) and Tanimoto coefficient-based methods.

Main Results:

• Both BIN and BBN methods enabled effective virtual screening searches.
• BBN demonstrated capability in handling diverse molecular datasets.
• Performance of BBN and BIN was comparable to simpler Tanimoto coefficient methods.

Conclusions:

• BBN offers an alternative approach for virtual screening, particularly for complex datasets.
• The effectiveness of BBN warrants further investigation in drug discovery pipelines.
• Simpler methods may offer comparable performance to advanced Bayesian networks for certain screening tasks.