Improving Protein-Ligand Docking Results with High-Throughput Molecular Dynamics Simulations

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Structure-based virtual screening uses scoring functions that struggle to differentiate binders from non-binders.
• Accurate prediction of protein-ligand interactions is crucial for efficient drug discovery.

Purpose of the Study:

• To improve the accuracy of virtual screening by integrating molecular dynamics (MD) simulations with docking results.
• To evaluate the effectiveness of MD simulations in distinguishing active ligands from decoys.

Main Methods:

• High-throughput protein-ligand complex MD simulations were performed using AutoDock Vina docking outputs.
• Ligand-binding stability was assessed by analyzing root-mean-square deviations from MD trajectories.
• A dataset of 56 protein targets and 560 ligands (280 actives, 280 decoys) from the DUD-E dataset was used.

Main Results:

• A 22% improvement in ROC AUC was achieved, increasing from 0.68 (AutoDock Vina alone) to 0.83.
• The MD simulation approach demonstrated robust performance across seven different protein classes.
• MD simulations moderately refined some predicted ligand-binding modes.

Conclusions:

• Physics-based MD simulations provide a reliable method for evaluating protein-ligand binding interactions.
• Integrating MD simulations with docking significantly enhances virtual screening performance.
• This approach offers a more accurate way to identify potential drug candidates.