What Molecular Docking Is (and Isn't)

Molecular docking is a computational method that predicts how a small-molecule ligand binds inside a protein receptor. Concretely, it estimates two things: the bound pose, meaning the ligand's 3D position, orientation, and internal conformation in the binding site, and a score that approximates binding affinity. You give it a receptor structure and a ligand, and it returns candidate poses ranked by that score.

Where Docking Fits

Docking is the workhorse of structure-based drug discovery, where you already have a 3D receptor structure from X-ray crystallography, cryo-EM, NMR, or a predicted model such as AlphaFold. Typical uses are virtual screening of large compound libraries to prioritize which molecules to test, and generating a plausible binding hypothesis to guide medicinal chemistry. It is a filter and a hypothesis generator, not a replacement for a wet-lab assay.

Two Coupled Problems

Every docking run solves two problems at once. The search problem samples many ligand poses by varying its translation, rotation, and rotatable-bond torsions inside the pocket. The scoring problem assigns each sampled pose an approximate binding energy so the poses can be ranked. The command below runs AutoDock Vina, where exhaustiveness controls the search effort and the output affinities come from the scoring function.

vina \
  --receptor receptor.pdbqt \
  --ligand ligand.pdbqt \
  --center_x 11.0 --center_y 90.5 --center_z 57.0 \
  --size_x 20 --size_y 20 --size_z 20 \
  --exhaustiveness 8 \
  --seed 42 \
  --out docked_poses.pdbqt

mode |   affinity | dist from best mode
     | (kcal/mol) | rmsd l.b.| rmsd u.b.
-----+------------+----------+----------
   1       -9.2       0.000      0.000
   2       -8.7       1.943      2.771
   3       -8.5       2.014      3.102
   4       -7.9       3.117      5.264
   5       -7.6       2.880      4.951

Each row is one pose the search produced, and the affinity column is the score used to rank them, in kcal/mol, where more negative is a stronger predicted interaction. The RMSD columns just measure how far each pose sits from the top-ranked one. Mode 1 is the program's best guess at the binding pose, but ranking first is a prediction, not a measurement.

The Rigid-Receptor Assumption

Most standard docking treats the ligand as flexible and the receptor as rigid. The ligand explores its rotatable bonds plus overall position and orientation, while the protein is frozen in the single conformation you supplied. This flexible-ligand, rigid-receptor model keeps the search tractable, but it means induced-fit motion, backbone rearrangement, and most side-chain flexibility are ignored unless you explicitly enable a limited set of flexible residues.