Ligand-based virtual screening using Bayesian inference network and reweighted fragments.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Traditional similarity-based virtual screening methods often assign equal importance to all molecular fragments, regardless of their relevance to biological activity.
• Bayesian networks offer an alternative approach to virtual screening by enabling more nuanced analysis of molecular structures.

Purpose of the Study:

• To improve the retrieval performance of similarity-based virtual screening by reweighting molecular fragments.
• To investigate the effectiveness of using active reference structures for fragment reweighting within a Bayesian inference network (BIN).

Main Methods:

• Employed Bayesian inference networks (BINs) for similarity-based virtual screening.
• Developed a fragment reweighting strategy using a set of active reference structures.
• Assigned higher weights to frequently occurring fragments in active structures and penalized less frequent ones.

Main Results:

• The proposed fragment reweighting approach significantly improved the retrieval effectiveness of ligand-based virtual screening.
• Performance gains were particularly notable when screening for structurally diverse active molecules.
• Simulated virtual screening experiments on MDL Drug Data Report datasets validated the method's efficacy.

Conclusions:

• Reweighting molecular fragments based on their frequency in active reference structures enhances BIN-based virtual screening.
• This method offers a significant improvement over existing tools, especially for structurally heterogeneous targets.
• The approach provides a more effective strategy for identifying potential drug candidates in virtual screening campaigns.