How to Use GNINA

GNINA is an open protein-ligand docking engine that combines AutoDock Vina-style pose search with convolutional neural-network scoring. The GNINA paper showed that CNN-based rescoring improves pose discrimination and makes docking outputs more useful for virtual screening than classical empirical energy terms alone.

On Neurosnap, researchers upload an Input Receptor structure and an Input Ligand and receive a ranked pose set for binding-mode inspection, hit triage, and downstream rescoring. The workflow is especially useful when a project needs more than one score: docking energy, ligand strain, pose plausibility, learned affinity-style ranking, and model uncertainty can all be compared together.

This makes GNINA a practical front-end for structure-based drug discovery, fragment follow-up, and ligand-series prioritization when the next question is which complexes deserve medicinal-chemistry attention or deeper simulation.

GNINA searches translational, rotational, and conformational ligand space using the Vina framework, then applies CNN ensembles to evaluate whether a pose looks like a realistic binding mode. That hybrid design matters scientifically because empirical docking scores and learned 3D interaction patterns capture different kinds of signal; the paper showed that using them together improves pose ranking and screening performance.

On Neurosnap, the receptor-ligand setup stays close to the way docking studies are actually run: one prepared receptor, one ligand, immediate pose ranking, and browser-side inspection of the proposed complex. The ranking table is most informative when read comparatively across affinity, intramolecular strain, CNN pose score, CNN affinity-style score, virtual-screening score, and CNN prediction variance.

Researchers generally use GNINA to narrow a hypothesis set rather than to declare a definitive binder. Strong candidates are the poses that look chemically plausible in the viewer and remain competitive across both classical and learned metrics.

• Direct docking workflow: GNINA starts from a receptor structure and ligand, which matches the standard setup for structure-based screening.
• Hybrid scoring: Classical docking search is reinforced with learned 3D interaction scoring, improving pose ranking and hit triage.
• Decision-ready metrics: Affinity, strain, CNN pose quality, CNN virtual-screening score, and variance can be interpreted together instead of relying on one number.
• Fast structural review: Ranked poses can be inspected immediately in the viewer before rescoring, medicinal-chemistry review, or experiment.