Molecular docking

What is molecular docking?

Molecular docking is a computational method for generating and ranking candidate bound arrangements of two molecular partners. The partners may be a small molecule and a protein, two proteins or a peptide and a receptor. A docking result is a structural hypothesis, not a direct observation of binding.

Every docking calculation combines a search with a way to rank the structures it finds. The search may move a ligand through a defined pocket, survey an entire protein surface or adjust both partners. Scoring may use approximate interaction energies, learned patterns or model confidence, each with a different meaning.

The right method follows the molecular system and the evidence available. A small-molecule engine cannot simply substitute for a protein interface method, and a good score cannot compensate for a wrong protonation state, an inaccessible surface or an omitted binding partner.

When to use molecular docking

Docking is useful when a concrete structural question can guide preparation and review:

• Propose contacts for an interaction supported by other evidence.
• Compare plausible orientations in a known site or interface.
• Test sensitivity to receptor state or docking method.
• Choose models for mutagenesis, simulation or structural follow-up.

Strengths and limitations

Docking returns explicit three-dimensional complexes that can be inspected and challenged. It can show which residues a ligand might contact, where alternative orientations arise and how a conclusion changes with a different receptor state or search method.

The search samples only part of a large conformational space. Preparation and biological context can dominate the answer, while fast scoring functions omit important solvent and conformational effects. Even a chemically plausible, high-ranked pose does not establish affinity, occupancy or function.

Which molecular docking tool to use

Two things decide the tool: what the partners are, and whether you already know where they bind. Every tool below runs online from an uploaded PDB, SDF or SMILES file and quotes its credit cost before it starts. An AutoDock Vina run starts at 15 credits.

What you want to do	Tool	What you get back
Dock a small molecule into a known pocket	AutoDock Vina	Ranked poses with a Vina score in kcal/mol
Dock into a known pocket and rescore with a neural network	GNINA	Ranked poses with CNN and empirical scores
Dock a ligand when the pocket is unknown	DiffDock-L	Ranked poses with learned confidence, no box needed
Let the pocket move as the ligand binds	DynamicBind	Poses with a ligand-adapted protein conformation
Find candidate pockets before docking	fpocket	Pockets ranked by druggability, with geometric descriptors
Dock two proteins or a protein and a peptide	LightDock	Ranked complex models from a swarm search
Dock two proteins with a learned model	DFMDock	Diffusion-sampled complexes ranked by learned energy
Check that a pose is physically sensible	PoseBusters	Pass or fail on bond geometry, clashes and stereochemistry
List the contacts in a pose	ProLIF	Residue-level interaction fingerprints

AutoDock Vina is the usual starting point for a small molecule in a known site, because its scores are widely reported and easy to compare with published work. DiffDock-L suits a ligand with no defined pocket, but its confidence ranks its own poses and is not a binding energy. Whichever engine you use, run PDBFixer on the receptor first if the structure has missing atoms or residues.

What a molecular docking calculation actually does

The inputs define which molecules and conformations exist in the calculation. The search then proposes relative placements and, where supported, changes in ligand torsions or protein geometry. The ranking model orders those proposals. A run can fail by missing the right configuration, scoring it poorly or starting from a structure in which that configuration is inaccessible.

The model family determines what its numbers mean. AutoDock Vina reports an approximate docking energy in kcal/mol. GNINA adds convolutional-network scores. DiffDock-L returns learned pose confidence. These numbers answer different questions and should remain attached to their native methods.

How to choose a molecular docking method

First identify the partners. Protein–ligand docking handles a small molecule in a protein site, and is the case the tools above are built for. Protein–protein docking addresses broad macromolecular interfaces. Antibody–antigen docking adds paratope and epitope context, while peptide–protein docking must contend with a flexible peptide backbone.

Then decide what is uncertain. Blind docking searches when the site is unknown. Flexible molecular docking permits receptor motion during pose prediction, whereas ensemble docking tests several prepared receptor states independently. Consensus docking compares methods under a stated agreement rule. AI molecular docking uses learned generative or hybrid models; its confidence remains model-specific.

Question	Starting approach	Main boundary
Small molecule, supported pocket	Local protein–ligand docking	The defined search region excludes other sites
Unknown site	Blind docking	Broad search increases false positives
Two macromolecular partners	Partner-specific complex docking	Interface and conformational uncertainty dominate
Several credible receptor states	Ensemble docking	Scores do not measure state populations
Uncertain scoring or search	Consensus comparison	Agreement can reflect shared bias

Applications of molecular docking

In structure-guided research, a docked complex can suggest which contacts to test by mutation or which compound changes might alter a binding orientation. A redocking control against a known complex tests the protocol's ability to recover an observed pose on that system, though it does not guarantee success for new chemistry.

For poorly characterized interactions, docking can organize sparse evidence into competing models. The most useful result may be a disagreement: two methods favor different interfaces, or a pose appears only in one receptor conformation. Those dependencies help design experiments that distinguish hypotheses.

How to do molecular docking online

1. Classify the partners. Choose the matching ProteinIQ workflow for a small molecule, protein partner, antibody or peptide.
2. Prepare the structures. Check chain identity, missing atoms, cofactors, waters, ligand stereochemistry and protonation. PDBFixer can repair justified structure gaps.
3. Define the search. Supply a supported pocket, interface restraints or receptor states before running; record what evidence set those boundaries.
4. Run the matched method. The featured method panel compares AutoDock Vina, GNINA, DiffDock-L and DynamicBind for a protein and small molecule. A macromolecular question needs its partner-specific workflow instead.
5. Review native outputs. Keep poses, scores, confidence values and logs by tool, and reject impossible geometry with PoseBusters where applicable.
6. Test the hypothesis. Compare with known structures, mutagenesis or binding measurements according to the question.

How to interpret molecular docking results

Start with input validity and pose geometry, then examine whether the proposed contacts fit known biology. Compare scores within their own method and preparation protocol. A Vina energy, a CNN score and diffusion confidence cannot be combined into one affinity estimate.

Preserve failed runs and alternative poses alongside the selected model. Docking can prioritize experiments or further calculations, but it does not confirm binding, activity, safety or function.

How the featured workflow works

A defensible workflow prepares inputs consistently, chooses a scientifically matched search strategy, runs one or more methods, and reviews geometry and uncertainty before validation.

1. Classify the system. Identify the partner class and biological question.
2. Prepare structures. Repair structures and standardize ligand chemistry.
3. Define the search. Set site, restraint, flexibility, or ensemble boundaries.
4. Generate models. Run complementary methods and preserve native outputs.
5. Validate. Inspect geometry and compare with orthogonal evidence.

Inputs and outputs for the featured workflow

Check formats before running, then inspect and download the result from every workflow step.

Inputs

Molecular structures

PDBSDFSMILES

Prepared structures appropriate to the selected partner class.

Search evidence

Binding sites, restraints, reference complexes, receptor ensembles, or known interactions when available.

Outputs

Candidate complexes

PDBPDBQTSDF

Method-native ranked poses or complex models.

Review package

Scores, confidence, contacts, logs, failures, provenance, and downloadable files.