Structure- and Ligand-Based Virtual Screening on DUD-E⁺: Performance Dependence on Approximations to the Binding...

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Virtual screening is crucial for identifying potential drug candidates.
• The accuracy of virtual screening depends heavily on the representation of the target protein and screening ligands.
• Traditional methods often rely on single protein structures or ligands, which may not capture the full binding site variability.

Purpose of the Study:

• To evaluate the impact of using single versus ensemble representations of protein pockets and ligands in virtual screening.
• To compare the performance of structure-based docking and ligand-based similarity screening.
• To investigate the efficacy of combining different virtual screening approaches for enhanced drug discovery.

Main Methods:

• Utilized the DUD-E+ benchmark dataset for virtual screening experiments.
• Employed structure-based docking (Surflex-Dock) and ligand-based screening (eSim) with single and ensemble representations.
• Assessed screening performance using metrics like ROC area and early enrichment.

Main Results:

• Ensemble approaches for both protein pockets and ligands significantly improved screening performance compared to single representations.
• Structure-based docking with ensembles yielded a mean ROC area of 0.84 ± 0.09, outperforming single-structure docking (0.81 ± 0.11).
• A hybrid approach combining ensemble docking and eSim screening achieved the highest performance (mean ROC area of 0.89 ± 0.08).

Conclusions:

• Relying on a single protein structure or ligand for virtual screening is suboptimal and can lead to reduced efficiency.
• Ensemble-based strategies for both structure- and ligand-based methods enhance screening accuracy.
• Combining complementary structure- and ligand-based approaches, particularly ensemble docking with eSim, offers a powerful strategy for improving drug discovery.