Condorcet and borda count fusion method for ligand-based virtual screening.

Area of Science:

• Computational chemistry
• Cheminformatics

Background:

• Individual similarity measures in virtual screening may not universally optimize active molecule structure recall across all activity classes.
• Data fusion, specifically similarity fusion, offers a method to enhance ligand-based virtual screening effectiveness by integrating multiple similarity measures.
• Similarity fusion combines similarity scores or rankings from various measures to produce a final compound database ranking.

Purpose of the Study:

• To evaluate the Condorcet fusion method for improving ligand-based virtual screening.
• To determine if combining multiple similarity measures via Condorcet fusion enhances the identification of active molecules.
• To assess the performance of Condorcet fusion across different data sets and molecular activity classes.

Main Methods:

• The Condorcet fusion method was applied, integrating outputs from eleven association and distance similarity coefficients.
• The best performing similarity measure for each molecule class was identified using Condorcet fusion.
• Performance was evaluated using the recall of active molecules at the top 5% and statistical significance tests.

Main Results:

• The Condorcet fusion method successfully combined outputs from eleven similarity coefficients.
• The method demonstrated improved recall of active molecules, particularly for sets with lower structural heterogeneity.
• Performance gains were modest for highly diverse activity sets.

Conclusions:

• Condorcet fusion offers a straightforward approach to enhance ligand-based virtual screening.
• The method is most effective when seeking active molecules with limited structural diversity.
• Further investigation may be needed to optimize fusion strategies for highly diverse molecular activities.