Use GNINA Online
Commercially Available GNINA No-Code Web Server
GNINA: Deep Learning Molecular Docking and Protein-Ligand Pose Scoring
GNINA is an open-source molecular docking program that uses an ensemble of convolutional neural networks (CNNs) to score protein-ligand poses. GNINA 1.0 was described by Andrew McNutt, Paul Francoeur, Rishal Aggarwal, Tomohide Masuda, Rocco Meli, Matthew Ragoza, Jocelyn Sunseri and David Ryan Koes (University of Pittsburgh), among others, in the Journal of Cheminformatics in 2021. Its code is available on GitHub.
Structure-based drug design depends on predicting where and how a small molecule binds a protein. Classical scoring functions such as the one in AutoDock Vina are fast, but they often rank near-native poses poorly. GNINA forks Smina, which in turn forks AutoDock Vina, and adds CNN scoring so that sampled poses are ranked by a learned model of the binding site rather than by a hand-built energy function alone.
Property | Detail |
|---|---|
Method | Monte Carlo pose sampling plus CNN ensemble scoring |
Outputs per pose | CNNscore (pose quality) and CNNaffinity (predicted pK) |
Default ensemble | Five CNN models, applied as rescoring |
Default settings | Exhaustiveness 8, 50 saved poses, 9 output modes |
Venue | Journal of Cheminformatics, 2021, vol. 13, article 43 |
How GNINA Docks and Scores Ligands with CNN Ensembles
GNINA keeps the Vina-style search and replaces the final judgment with deep learning.
Define the search space: a Cartesian box, or a box built around a reference ligand and expanded by a configurable margin (default 4 Angstroms).
Sample poses: Monte Carlo chains apply translations, rotations and torsion changes, followed by energy minimization.
Rescore with CNNs: the default five-model ensemble rates each pose. Poses are sorted by CNNscore by default.
Estimate affinity: each CNN also outputs a CNNaffinity value in pK units.
Optional Flexible Receptor Docking
Side-chain torsions can be sampled while the backbone stays fixed. Flexible residues can be chosen from a file, a residue list or by distance from the ligand.
GNINA Benchmark Results: Redocking, Cross-Docking and Whole-Protein Docking
The authors compared GNINA with AutoDock Vina on redocking (PDBbind, 4,260 protein-ligand pairs) and cross-docking (7,970 pairs across 92 pockets). Top1 success means a pose under 2 Angstroms RMSD is ranked first.
Setting | Vina Top1 | GNINA Top1 |
|---|---|---|
Redocking, pocket defined | 58% | 73% |
Cross-docking, pocket defined | 27% | 37% |
Redocking, whole protein | 31% | 38% |
Cross-docking, whole protein | 12% | 16% |
On held-out systems unseen in training (441 redocking and 178 cross-docking pairs), the default ensemble reached 68% and 42% Top1, against about 57% and 23% for Vina. CNNscore is also informative: 87% of redocking systems had a top pose scoring above 0.8, and higher scores generally tracked lower RMSD.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with GNINA on Tamarind Bio
Virtual screening and hit finding: dock large ligand sets to a defined pocket and rank them by CNNscore and CNNaffinity.
Pose prediction: redock or cross-dock known ligands to propose binding modes for a target.
Blind docking: search the whole protein surface when the binding site is unknown, accepting lower accuracy.
Pose triage: use CNNscore as a confidence signal to decide which predicted poses deserve follow-up.
How to Use GNINA on Tamarind Bio
Open the tool: log in to tamarind.bio and open GNINA.
Provide the receptor: upload the protein structure, for example a PDB file.
Provide the ligand: supply the small molecule in a supported format such as SDF, if exposed.
Define the search region: set a box or a reference ligand for pocket-defined docking, or leave it open for whole-protein docking if exposed.
Set parameters: adjust options such as exhaustiveness (default 8) or the number of output poses, if exposed.
Run the job: submit and wait for docking to finish.
Download and interpret: retrieve the poses and review CNNscore and CNNaffinity for each.
Use downstream: carry top poses into visual inspection, further scoring or compound prioritization.
Things to Keep in Mind
Whole-protein docking is much less accurate than pocket-defined docking because ligands can get stuck in local minima away from the true site.
Confidence is lower for cross-docking: only 15% of cross-docked ligands had a top pose with CNNscore above 0.8.
Flexible docking showed no clear advantage over rigid docking, and the default CNNs were not trained for it.
CNN refinement is costly and was not found to beat rescoring.
Speed depends on GPU availability.
Source: McNutt, A.T., Francoeur, P., Aggarwal, R., Masuda, T., Meli, R., Ragoza, M., Sunseri, J. and Koes, D.R. (and others), GNINA 1.0: molecular docking with deep learning. Journal of Cheminformatics 13, 43 (2021). DOI: 10.1186/s13321-021-00522-2. Code: https://github.com/gnina/gnina