AutoDock Vina

What is AutoDock Vina?

AutoDock Vina is one of the most widely-used open-source molecular docking programs in computational drug discovery. First published in 2010, Vina has been cited over 17,000 times and is considered the gold standard for physics-based molecular docking. Unlike AI-based methods, Vina uses classical force fields and efficient search algorithms to predict how small molecules bind to protein targets.

How does AutoDock Vina work?

AutoDock Vina uses a physics-based approach to molecular docking:

Scoring function: Vina uses an empirical scoring function that estimates binding affinity based on:

• Intermolecular interactions (hydrogen bonds, hydrophobic contacts)
• Steric clashes and van der Waals forces
• Entropic penalties for conformational changes
• Solvation/desolvation effects

Search algorithm: Vina employs an iterated local search global optimizer:

• Broyden-Fletcher-Goldfarb-Shanno (BFGS) local optimization
• Monte Carlo sampling for global exploration
• Parallelized across multiple CPU cores for efficiency

Output: Binding affinity in kcal/mol - a thermodynamic measure of binding strength.

This classical approach is:

• Fast and computationally efficient (CPU-only, 1-2 minutes typical)
• Interpretable (direct physical meaning)
• Validated across thousands of protein-ligand systems

Input requirements

AutoDock Vina requires two inputs:

Protein (Receptor): A PDB file or PDB ID from RCSB PDB database. The protein is automatically converted to PDBQT format (PDB with partial charges and atom types). For best results:

• Remove water molecules unless critical for binding
• Ensure proper protonation at physiological pH
• Use tools like PDB Fixer for preparation

Ligand: SMILES string or SDF file. The ligand is converted to PDBQT format with:

• 3D coordinates generated (for SMILES)
• Hydrogens added
• Gasteiger partial charges assigned
• Rotatable bonds identified for flexible docking

Search box: Defines the 3D region where Vina searches for binding:

• Auto mode (recommended): Searches the entire protein using automatically calculated box
• Manual mode: Specify exact coordinates if you know the binding site (e.g., from literature or co-crystal structures)

Understanding the results

Vina generates multiple binding poses (modes) ranked by predicted binding affinity:

Mode: Pose number (Mode 1 = best affinity)

Binding Affinity (kcal/mol): Free energy of binding. More negative = stronger binding:

• -4 to -6 kcal/mol: Weak binding (mM to high μM range)
• -6 to -8 kcal/mol: Moderate binding (low μM range)
• -8 to -10 kcal/mol: Strong binding (nM range)
• < -10 kcal/mol: Very strong binding (sub-nM range)
• < -12 kcal/mol: Unusually strong - validate for potential artifacts

RMSD (Root Mean Square Deviation): Structural similarity between poses:

• rmsd_lb: Lower bound RMSD between binding modes
• rmsd_ub: Upper bound RMSD between binding modes
• Poses with RMSD < 2Å are considered similar

Configurable parameters

Exhaustiveness

Exhaustiveness controls the number of independent docking runs performed. Each run consists of sequential steps where the algorithm generates a random ligand conformation, performs local optimization using the Broyden-Fletcher-Goldfarsh-Shanno (BFGS) algorithm, and applies selection criteria.

• Default: 8 — Adequate for preliminary screening
• Quick tests: 4 — Fast exploration for initial assessments
• Critical work: 16-32 — Publication-quality results with high confidence
• Scaling behavior:
  • Runtime increases linearly (exhaustiveness 16 takes ~2× longer than 8)
  • Probability of missing global minimum decreases exponentially
• Parallel execution: Individual runs execute in parallel, making this the primary parameter for search thoroughness

Higher exhaustiveness ensures the algorithm thoroughly explores conformational space and increases confidence that the true global minimum (best binding pose) has been found rather than a local minimum.

Search Space Settings

The search space defines the 3D rectangular box where Vina searches for binding poses. Specified as center coordinates (X, Y, Z) and dimensions (size X, Y, Z) in Angstroms.

Critical constraints:

• Size limit: 30×30×30 Å — Stay under this volume for efficient searching
• Must contain binding site — Box too small misses correct pose; too large slows search dramatically
• Smaller = faster — Compact search spaces enable more thorough exploration within the same computational budget
• When exceeding 30×30×30 Å: Proportionally increase exhaustiveness to maintain search quality

Note: AutoDock Vina uses Angstroms directly (unlike AutoDock 4's grid points of 0.375 Å). The algorithm performs best with focused search spaces because it can explore a smaller volume more thoroughly.

Energy Range

Energy range (kcal/mol) filters which binding modes appear in results. Only poses within this energy difference from the best (lowest energy) mode are returned.

• Default: 3 kcal/mol — Returns poses up to 3 kcal/mol worse than the best
• Lower values (1-2): Fewer alternative poses, only near-optimal binding modes
• Higher values (5-10): More diverse binding modes, including suboptimal poses
• Effect: Controls output diversity—3 kcal/mol captures biologically relevant binding modes while filtering unlikely high-energy conformations

This parameter doesn't affect the search itself, only which results are shown.

Number of Poses

Number of poses (1-20, default: 9):

• Maximum binding modes to generate and return
• More poses provide better coverage of binding space
• Top 3-5 poses typically sufficient for analysis
• All poses must satisfy the energy range criterion

Best practices

Search Space Selection

• Auto mode: Use when binding site is unknown—searches entire protein with automatically calculated box
• Manual mode: Use for focused search when site is known (from literature, co-crystal structures, or computational predictions)
  • Faster execution and more accurate results
  • Keep total volume under 30×30×30 Å for optimal performance
  • 20-25Å per dimension typically sufficient for most binding pockets
  • If using larger boxes (>30Å), increase exhaustiveness proportionally
• Avoid oversized boxes (>50Å): Dramatically reduces accuracy and wastes computational resources

Exhaustiveness Settings

• Default (8): Adequate for preliminary screening and workflow development
• Production work (16): Recommended for lead optimization and comparative studies
• Critical applications (24-32): Use for publication-quality results or when confidence is paramount
• Quick tests (4): Only for rapid prototyping—insufficient for decision-making

Number of Poses

• Default (9) works well for most cases
• Increase to 15-20 for comprehensive binding mode exploration
• Always examine top 3-5 poses, not just the best—consider consensus across multiple modes
• Verify poses satisfy the energy range criterion (default: within 3 kcal/mol of best)

Result Validation

• Visual inspection: Check protein-ligand interactions (H-bonds, hydrophobic contacts, salt bridges)
• Experimental correlation: Compare with known binding data when available
• Statistical confidence: Consider running multiple independent docking runs with different random seeds
• Scoring limitations: Binding affinity is computational estimate—validate critical findings experimentally
• Pose clustering: Group similar poses (RMSD < 2Å) to identify consensus binding modes

Known Limitations

• Rigid receptor: Protein treated as rigid (no induced-fit effects)
• Scoring accuracy: Function accuracy varies by protein family and ligand type
• Large ligands: May struggle with macrocycles, peptides, or highly flexible molecules (>15 rotatable bonds)
• Metal coordination: Requires specialized parameters for metalloproteins
• Covalent docking: Not designed for covalent inhibitors

AutoDock Vina vs. DiffDock-L

Both tools are available on ProteinIQ for complementary workflows:

Recommendation: Use Vina for fast screening, known binding sites, and when physical interpretability matters. Use DiffDock-L for challenging targets, when accuracy is critical, or when exploring diverse binding modes.

Cost

Using AutoDock Vina through ProteinIQ costs 25 credits per docking job.

---

Based on: Trott, O., & Olson, A. J. (2010). AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. Journal of Computational Chemistry, 31(2), 455-461. DOI: 10.1002/jcc.21334