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

Drug discovery is an intricate and resource-intensive process in which computational approaches, such as molecular docking, are essential, particularly in the early stages, to identify potential hits. However, docking still has many drawbacks, including problems in managing protein flexibility and the reliability of scoring functions. In this paper, we systematically compared the performance of AutoDock Vina, one of the most widely used open-source docking tools, with GNINA. This advanced evolution integrates convolutional neural networks (CNNs) for pose scoring. The comparison was conducted on ten heterogeneous protein targets, including metalloenzymes, kinases, and G-protein-coupled receptors (GPCRs). With the ability to accurately replicate binding poses and their energy values, GNINA showed outstanding performance in both virtual screening (VS) of active ligands and re-docking steps of co-crystallized ligands. GNINA’s enhanced ability to accurately distinguish between true positives and false positives—a specificity not found with AutoDock Vina—is confirmed by ROC curves and Enrichment Factor (EF) results. Therefore, we propose an integrated GNINA-based workflow that can significantly enhance the quality and reliability of docking results, providing a valuable tool for optimizing the initial stages of drug discovery.