Optimal assignment methods for ligand-based virtual screening.

Area of Science:

• Computational chemistry
• Drug discovery
• Cheminformatics

Background:

• Ligand-based virtual screening is crucial for early drug discovery, aiming to identify novel chemical scaffolds.
• Existing similarity methods struggle with scaffold-hopping, limiting the discovery of diverse active molecules.
• The optimal assignment approach on molecular graphs, successful in QSAR, remains untested for virtual screening.

Purpose of the Study:

• To evaluate existing and novel optimal assignment methods for ligand-based virtual screening.
• To assess the scaffold-hopping capabilities of these methods using chemotype clustering.
• To compare their early recognition performance against literature benchmarks.

Main Methods:

• Evaluation of two published and two new optimal assignment methods.
• Utilizing chemotype clustering information for performance metrics.
• Comparative analysis against established literature results.

Main Results:

• Improved early recognition performance in virtual screening compared to literature.
• Comparable overall dataset results with enhanced scaffold-hopping ability.
• A novel two-step assignment method demonstrated superior scaffold-hopping and early recognition.

Conclusions:

• Optimal assignment methods effectively combine chemotype discovery with active structure enrichment.
• The approach allows for interpretable mappings and parameter modification for target-specific tuning.
• Low computational times make these methods suitable for large-scale virtual screening.