Virtual screening applications: a study of ligand-based methods and different structure representations in four...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Ligand-based virtual screening (LBVS) is crucial for prioritizing compounds in drug discovery.
• Evaluating different LBVS strategies and molecular representations is essential for optimizing screening efficiency.

Purpose of the Study:

• To investigate the performance of two LBVS techniques: similarity search with data fusion (SSDF) and novelty detection with Self-Organizing Maps (ndSOM).
• To compare three molecular structure representations for their effectiveness in LBVS.
• To assess these methods across four distinct virtual screening scenarios.

Main Methods:

• Retrospective LBVS was performed for eight biological targets using the MDDR and WOMBAT databases.
• SSDF and ndSOM were evaluated using Daylight fingerprints, topological autocorrelation, and radial distribution functions.
• Performance was analyzed across four scenarios: HTS prioritization, small set selection, activity probability assessment, and assay-guided selection.

Main Results:

• Both SSDF and ndSOM were applicable for prioritizing compounds for high-throughput screening (scenario 1).
• Similarity search performed slightly better for selecting small sets of active compounds (scenario 2).
• Novelty detection with SOMs was preferred for assessing activity probability (scenario 3), while no method succeeded in scenario 4.

Conclusions:

• The choice of LBVS method and molecular representation depends significantly on the specific screening objective.
• SSDF and ndSOM offer complementary strengths for different stages of the virtual screening pipeline.
• Further development is needed for methods aiming to identify the most active compounds for biological assays.