High-Throughput, High-Quality: Benchmarking GNINA and AutoDock Vina for Precision Virtual Screening Workflow

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Molecular docking is crucial for early-stage drug discovery but faces challenges with protein flexibility and scoring function reliability.
• AutoDock Vina is a widely used open-source docking tool, yet its performance can be limited.

Purpose of the Study:

• To systematically compare the performance of AutoDock Vina with GNINA, a deep learning-based docking tool.
• To evaluate their efficacy in virtual screening and pose prediction for drug discovery.

Main Methods:

• Comparison of AutoDock Vina and GNINA on ten diverse protein targets (metalloenzymes, kinases, GPCRs).
• GNINA utilizes convolutional neural networks (CNNs) for enhanced pose scoring.
• Performance assessed using virtual screening of active ligands and re-docking of co-crystallized ligands.

Main Results:

• GNINA demonstrated superior performance in accurately replicating binding poses and energy values compared to AutoDock Vina.
• GNINA showed enhanced specificity in distinguishing true positives from false positives, confirmed by ROC curves and Enrichment Factor (EF).
• GNINA excelled in both virtual screening and re-docking tasks.

Conclusions:

• GNINA offers improved accuracy and reliability in molecular docking for drug discovery.
• A GNINA-based workflow can significantly enhance the quality of docking results in early-stage drug development.
• GNINA represents a valuable advancement for optimizing hit identification in drug discovery pipelines.