Skip to content

Concepts

A short primer on what PocketDock does and the terminology it uses. Skip this page if you're already comfortable with structure-based virtual screening.

Molecular docking, in one paragraph

Docking predicts how a small molecule (the ligand) binds to a protein (the receptor). The output is a set of candidate poses — each pose is a 3D arrangement of the ligand within a binding site, scored by an estimated binding affinity. Better-scoring poses are more likely to represent the true bound state. Docking does not simulate molecular dynamics; it samples plausible orientations and ranks them with a fast scoring function.

Binding pockets

Most drug-like ligands bind in cavities on the protein surface — clefts deep enough to accommodate a small molecule and lined with the right mix of hydrophobic and polar residues. These cavities are binding pockets.

PocketDock uses P2Rank to predict pockets directly from the 3D structure. P2Rank is a machine-learning model that scores patches of the protein surface by how druggable they look. For each predicted pocket you get:

  • Probability (0–1) — likelihood the pocket can bind a drug-like molecule.
  • P2Rank score — an internal pocket-quality metric (higher is better).
  • Center coordinates — used by Vina to position the docking grid.
  • Lining residues — the amino acids forming the pocket wall.

You don't need to specify a binding site in advance. PocketDock docks into the top N pockets (you set N — default is 3).

Docking poses

For each pocket, AutoDock Vina searches for ligand conformations and orientations that minimize the predicted binding free energy inside the docking box. It returns up to 9 ranked poses per pocket by default, ordered from best (lowest energy) to worst.

Each pose has:

  • Affinity in kcal/mol (more negative = stronger predicted binding).
  • RMSD lower bound and upper bound vs. the best pose — measures of pose uncertainty in Ångströms.

Binding affinity, in numbers

Vina's affinity is an estimate of the binding free energy ΔG. As a rough guide:

Affinity (kcal/mol) Predicted strength
≤ −10 Very strong
−10 to −8 Strong
−8 to −6 Moderate
−6 to −4 Weak
> −4 Very weak

These ranges are heuristics — Vina's scoring function is fast but approximate, so treat affinities as rankings rather than absolute predictions. Confidence is highest when comparing similar ligands against the same target.

Exhaustiveness

Vina's exhaustiveness parameter controls how many independent search runs are performed inside the docking box. Default is 8. Doubling it roughly doubles runtime but reduces the chance of missing a low-energy pose. Use 16 or 32 for tricky systems (flexible ligands, large pockets); use the default for screening.

What the "combined score" is

PocketDock reports a combined score in the results table that blends pocket probability with binding affinity:

combined = 0.4 × pocket_probability + 0.6 × normalize(affinity, max = -15.0)
  • pocket_probability is in [0, 1] from P2Rank.
  • normalize(affinity, max=-15.0) maps affinity to [0, 1] such that −15 kcal/mol = 1.0 and 0 kcal/mol = 0.0.

The intent: a pose ranks well only if it sits in a credible pocket and binds tightly. See Interpreting Results for a worked example.

  • Ligand efficiency (LE) — affinity normalized by ligand size; lets you compare a tight-binding small molecule against a tight-binding large molecule. PocketDock displays it in the pose info panel.
  • Dissociation constant (Kd) — affinity expressed as a concentration; PocketDock estimates it from the docking score using the Boltzmann relation. Useful as an intuitive number, but inherits all the uncertainty of the docking score.
  • Interactions — specific contacts between ligand and pocket residues (H-bonds, hydrophobic, salt bridges, π-stacking, π-cation, halogen). PocketDock detects six types geometrically and visualizes them in the 3D viewer. See Interaction Analysis.

Beyond single-ligand docking

PocketDock v2.0 adds several capabilities on top of the basic pocket+Vina flow. Each has its own user-guide page; the short definitions below are enough to orient yourself.

Batch job

A batch job is a single submission that creates one DockingJob per ligand against a shared protein. PocketDock supports up to 100 ligands per batch; multi-molecule SDF files are split automatically. All jobs in a batch share a batch_id and appear on a single dashboard at /batch/<batch_id>/. See Batch Docking.

Ensemble docking

Ensemble docking generates several plausible conformations of the receptor and docks the ligand into all of them. PocketDock supports two methods:

  • NMA (Normal Mode Analysis) — fast (~30 s for 5 conformations). Builds an anisotropic network model from the Cα atoms, then perturbs the structure along the lowest normal modes. Captures slow, collective backbone motions.
  • MD (short OpenMM molecular dynamics) — slower (5–15 min). Runs a 20 ps Langevin simulation at 300 K with AMBER14 + OBC2 implicit solvent and saves N evenly-spaced snapshots. Captures local relaxation including side-chain rearrangements.

The number of conformations (num_conformations) is configurable from 2 to 10. Each child conformation becomes its own docking job; the ensemble dashboard at /ensemble/<ensemble_id>/ aggregates them and shows a consensus top-20 pose ranking by combined score. See Ensemble Docking.

ADMET, Lipinski, Veber, QED

ADMET stands for Absorption, Distribution, Metabolism, Excretion, Toxicity — the in-vivo behaviors that determine whether a binder can become an actual drug. PocketDock doesn't predict ADMET directly; it computes a set of fast drug-likeness descriptors from RDKit that correlate with good oral-drug behavior.

  • Lipinski's rule of five — MW ≤ 500, logP ≤ 5, H-bond donors ≤ 5, H-bond acceptors ≤ 10. The count of broken rules is the Lipinski violation number.
  • Veber's rules — TPSA ≤ 140 Ų and rotatable bonds ≤ 10.
  • QED — Quantitative Estimate of Drug-likeness on [0, 1]. A single number that blends several Lipinski-style features and structural alerts. Marketed oral drugs typically sit in 0.5–0.8.

ADMET runs on every job automatically; the panel appears on the results page. See ADMET Properties.

Pose refinement

Pose refinement is an optional post-docking step that energy-minimizes each Vina pose with OpenMM (AMBER14-all force field + OBC2 implicit solvent, Langevin dynamics at 300 K, no constraints). The result is a slightly relaxed pose with reduced steric clashes and small geometry adjustments. After refinement, PocketDock re-runs the interaction analysis so the contact list reflects the relaxed geometry. Enabled by ticking Refine poses on the upload form.

MM-GBSA score

A second, physics-based binding-energy estimate added to each pose when MM-GBSA rescore is enabled. PocketDock's implementation is a lightweight stand-in for the full AMBER MM-GBSA protocol: it uses RDKit MMFF94 + Gasteiger charges to compute a per-pose interaction energy (Coulomb + Lennard-Jones) plus a ligand-strain term, reported in kJ/mol (more negative = stronger binding). It is not a rigorous ΔG prediction — use it as a second opinion to Vina, not as a replacement for it. See MM-GBSA Rescoring.

What PocketDock is not

  • Not a molecular dynamics simulator — there's no time evolution or solvation modeling.
  • Not a free-energy perturbation tool — the affinities are scoring-function estimates, not rigorous ΔG predictions.
  • Not a docking validator — predicted poses still need experimental confirmation for any high-stakes claim.

For most exploratory and ranking work — virtual screening, hypothesis generation, teaching — that's plenty.