On evaluating molecular-docking methods for pose prediction and enrichment factors

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Commercial molecular docking programs are essential tools in drug discovery for predicting ligand-protein interactions.
• Evaluating the performance of these programs is crucial for optimizing virtual screening strategies.

Purpose of the Study:

• To compare the ligand-docking and virtual-screening capabilities of four leading docking programs: FlexX, GOLD, GLIDE, and ICM.
• To assess their performance in reproducing native ligand conformations and enriching active compounds in virtual screening experiments.

Main Methods:

• The study evaluated 164 high-resolution protein-ligand complexes to assess the programs' ability to reproduce native ligand conformations from SMILES strings.
• Virtual screening was performed using the latest versions of the four programs and their native scoring functions against 12 therapeutic protein targets.
• Performance was measured by the programs' ability to rank target-specific active compounds over decoys and randomly selected compounds.

Main Results:

• The four docking programs exhibited varying performance in reproducing native ligand conformations.
• Virtual screening results showed differences in the programs' ability to enrich active compounds, with some outperforming others in specific target systems.
• Enrichment factors were compared against 3D shape-matching and 2D similarity database search methods.

Conclusions:

• No single docking program consistently outperformed the others across all tested metrics and targets.
• The choice of docking program and scoring function can significantly impact the success of virtual screening campaigns.
• Further optimization and validation of docking programs are needed to enhance their reliability in drug discovery pipelines.