AutoDock Vina

What it simulates

AutoDock Vina docks a small molecule (the ligand) into a protein (the receptor) and ranks the poses it finds. It is a search plus a score, with no dynamics. The receptor is rigid. The ligand can twist around its rotatable bonds. You give Vina a box, and it searches for ligand positions, orientations and twists inside that box.

Each pose gets a score in kcal/mol. More negative means a better predicted fit. Vina reports up to --num_modes poses and lists how far each differs from the best one as RMSD lower and upper bounds.

The key assumptions are a rigid receptor, a search limited to your box, and a scoring function fitted to known binding data. The score ranks poses. It does not give a measured binding energy.

Input is a receptor and a ligand, both in PDBQT format (PDB plus partial charges and atom types), and the box centre and size in Angstrom. Output is a PDBQT file with all poses and a log with the score table.

When to use it

• You have a protein structure and a few candidate ligands and want to rank them or see where they sit.
• You want to ask whether a known ligand's pose is reproduced (redocking).
• You want a quick screening step before more costly methods.

OpenMM and GROMACS on biosimhub fit when you need the motion of a complex over time, with flexible protein and water. Use Vina first to place the ligand, then a molecular dynamics tool to test whether the pose holds. Vina does not do dynamics.

Install it yourself

Pinned to 1.2.7 (conda-forge, checked with a micromamba dry run, 90 MB of downloads):

micromamba create -n vina -c conda-forge vina=1.2.7
micromamba activate vina

The Vina documentation prepares inputs with Meeko. The commands from its basic docking tutorial are:

mk_prepare_receptor.py -i 1iep_receptorH.pdb -o 1iep_receptor -p -v
mk_prepare_ligand.py -i 1iep_ligand.sdf -o 1iep_ligand.pdbqt

The tutorial prepares the ligand from an SDF file, since a PDB file has no bond information. Meeko is a separate package and is not in the recipe's image.

The vina command line, as printed by vina --help in the 1.2.7 container, has these main options: --receptor, --ligand, --center_x/y/z, --size_x/y/z, --exhaustiveness (default 8), --num_modes (default 9), --scoring (ad4, vina or vinardo, default vina), --seed (default 0), --cpu and --out.

Minimal run:

vina --receptor receptor.pdbqt --ligand ligand.pdbqt \
  --center_x 0 --center_y 0 --center_z 0 --size_x 14 --size_y 14 --size_z 14 \
  --out docked.pdbqt

Reading the output

The example is a toy. The receptor is 48 atoms placed around the origin. The ligand has 5 atoms (C1, C2, C3, O1 and HO) and one rotatable bond (TORSDOF 1). The example is a toy so that it runs in about a second. The scores are therefore weak. A real protein-ligand pair scores much lower.

The example runs vina directly with --cpu 1 and seed 42, with no helper script.

• vina_log.txt (2,068 bytes): the Vina banner and citations, then the settings: Scoring function : vina, Grid center: X 0 Y 0 Z 0, Grid size : X 14 Y 14 Z 14, Exhaustiveness: 8, CPU: 1 and Performing docking (random seed: 42). After the progress bar comes the table of 9 modes.
• docked.pdbqt (6,075 bytes): 9 MODEL blocks, one per pose. Each starts with REMARK VINA RESULT: -1.231 0.000 0.000, which holds the score and the two RMSD bounds. The next lines split the score into INTER + INTRA, INTER, INTRA and UNBOUND.

The best pose scores -1.231 kcal/mol (INTER -1.303, INTRA 0). The other eight range from -1.228 to -1.184. The poses differ by 2.2 to 6.7 Angstrom RMSD from the best, so the scores of very different poses are within 0.05 kcal/mol. The table is flat because the toy receptor does not favour one pose.

Your own runs

Same image and box; only the stated option changed.

Vinardo scoring (--scoring vinardo). The best pose scores -1.415 and the nine scores run from -1.415 to -1.383. The spread is 0.03 kcal/mol, against 0.05 for Vina scoring. Treat Vina and Vinardo values as separate scales.

More search effort (--exhaustiveness 32). The best score is -1.232, against -1.231 at 8, and the RMSDs of the other poses change (the 2nd pose sits 5.8 Angstrom from the best, against 2.2). The Vina tutorial says higher exhaustiveness gives a more consistent result. Here the top score barely moves because the toy surface is flat.

Different seed (--seed 5). The best score is -1.232, and the 2nd pose is -1.225 at 3.6 Angstrom, against -1.228 at 2.2 with seed 42. Seeds change which poses are reported, so compare seeds when the table is flat.

Box away from the pocket (--center_x 30). All nine poses score 0.000. The ligand has no receptor contact there.

Smaller box (--size_x 4 --size_y 4 --size_z 4). The best score is -0.936, and only eight modes come back instead of nine. A box that is too tight can exclude better poses.

Pitfalls

• The box decides the answer. Vina searches only the box. Place it on the binding site and make it a few Angstrom larger than the ligand. The Vina tutorial uses 20 x 20 x 20 Angstrom for its example.
• Seeds and exhaustiveness. The default seed is 0. Set one for reproducibility, and rerun with exhaustiveness 32 as in the tutorial before trusting a ranking.
• Score units. Scores are in kcal/mol, with negative values for favourable binding. Do not compare scores across scoring functions. The Vina tutorial says AutoDock and Vina scores are not comparable.
• PDBQT preparation. Vina needs charges and atom types in the PDBQT files. A plain PDB file does not work. Prepare the ligand from a file with bond information, such as SDF.
• Rigid receptor. Side chains do not move unless you supply --flex. Poses for ligands that need an induced fit can be wrong.

Further reading

• Documentation: https://autodock-vina.readthedocs.io/en/latest/
• Basic docking tutorial: https://autodock-vina.readthedocs.io/en/latest/docking_basic.html
• Source: https://github.com/ccsb-scripps/AutoDock-Vina
• Paper: Trott O, Olson AJ (2010), J Comput Chem 31:455-461, https://doi.org/10.1002/jcc.21334
• Vina 1.2 paper: Eberhardt J et al. (2021), J Chem Inf Model, https://doi.org/10.1021/acs.jcim.1c00203