Fusing similarity rankings in ligand-based virtual screening

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Ligand-based virtual screening (LBVS) is crucial for identifying potential drug candidates.
• Traditional LBVS relies on single similarity measures, which can be limiting.
• Combining multiple data sources, or data fusion, offers a promising approach to improve screening accuracy.

Purpose of the Study:

• To review data fusion methods for combining rankings in ligand-based virtual screening.
• To discuss similarity fusion and group fusion approaches.
• To examine rules for combining similarity rankings and evidence for fusion's superiority.

Main Methods:

• The review focuses on data fusion techniques applied to LBVS.
• Two primary fusion approaches are detailed: similarity fusion and group fusion.
• Analysis of rules for combining similarity rankings is presented.

Main Results:

• Similarity fusion combines rankings from multiple similarity measures.
• Group fusion combines rankings from multiple searches using a single measure.
• Evidence suggests fusion methods outperform conventional similarity searching.

Conclusions:

• Data fusion offers a robust strategy for enhancing ligand-based virtual screening.
• Both similarity and group fusion techniques provide advantages over single-measure approaches.
• Further investigation into fusion rules can optimize drug discovery pipelines.