Optimization and visualization of the edge weights in optimal assignment methods for virtual screening.

Area of Science:

• Cheminformatics
• Computational Chemistry
• Drug Discovery

Background:

• Ligand-based virtual screening is crucial in early drug discovery.
• Optimal assignment of chemical graphs is a common similarity function.
• Current methods assume all atoms are equally important, which is often unrealistic.

Purpose of the Study:

• To improve virtual screening by optimizing atom importance in optimal assignment.
• To integrate ligand binding information into molecular similarity calculations.
• To develop a visualization method for understanding atom importance.

Main Methods:

• Optimized edge weights in optimal assignment using evolutionary algorithms (differential evolution and particle swarm optimization).
• Evaluated performance using virtual screening metrics on a benchmark dataset.
• Developed a visualization technique to interpret optimized edge weights.

Main Results:

• Both evolutionary algorithms effectively optimized edge weights.
• The proposed method significantly outperformed standard optimal assignment and other similarity functions in overall and early enrichment.
• Visualizations revealed important substructures for ligand retrieval and protein binding.

Conclusions:

• Optimizing edge weights in optimal assignment is a valuable enhancement for ligand-based virtual screening.
• This approach is broadly applicable to similarity functions using optimal assignment.
• The visualization aids in understanding ligand-protein interactions and binding modes.