Protein–protein docking

What is protein–protein docking?

Protein–protein docking is a computational method for arranging two protein structures into plausible complexes. Methods differ in whether they use restraints, global rigid-body search, learned geometry, or flexible refinement.

Protein–protein docking searches the relative orientation of two large, irregular surfaces with thousands of possible contacts. Global methods explore broadly, while information-driven methods narrow the search with restraints, predicted interface residues, cross-links, mutagenesis, or known domain contacts. Some protocols then refine side chains or limited backbone motion around each candidate interface.

Every model remains conditional on the supplied conformations and evidence. Unbound structures may rearrange during association, flexible loops may be missing, and oligomeric state or domain orientation may be uncertain. Treat the output as a set of interface hypotheses to cluster, compare across methods, and test experimentally, not as one automatically correct complex.

When to use protein–protein docking

• Suitable protein–protein docking question. Building testable interaction-interface and complex-assembly hypotheses
• Required structures and evidence are available. Prepared structures or models for both partners, plus restraints or interface evidence when available

Benefits of protein–protein docking

• Practical output. Supports global and information-driven searches
• Comparative evidence. Produces explicit interface models
• Connected analysis. Makes method disagreement visible

Primary limitations

• Method dependence. Large conformational changes remain difficult
• Input sensitivity. False interfaces can score well
• Validation boundary. Benchmark metrics require a native reference

How protein–protein docking works

The appropriate search strategy depends on how much reliable interface information is available before docking.

• Global rigid-body docking. Methods such as LightDock and EquiDock explore orientations without requiring a known interface. They provide broad coverage but can generate many plausible-looking false interfaces.
• Information-driven docking. HADDOCK3 uses ambiguous or explicit interaction restraints to focus sampling and refinement. Restraints improve the search only when the underlying evidence is credible.
• Learned complex modeling. AF2Dock, DFMDock, GeoDock, and related approaches use learned structural priors. Their confidence and failure modes differ from energy-based docking and require independent review.

Applications of protein–protein docking

Protein–protein docking is used to convert interaction evidence into explicit structural hypotheses.

• Interface mapping. Identify candidate contact residues for mutagenesis, cross-linking, or construct design.
• Complex assembly. Model how known interaction partners or domains may assemble when no experimental complex structure is available.
• Method benchmarking. Compare global, restrained, and learned methods against a native reference using DockQ and interface-specific measures.

How to do protein–protein docking online

The ProteinIQ comparison workflow keeps both partner structures constant while running several protein–protein docking strategies side by side.

1. Prepare both protein partners. Upload separate PDB structures, select the correct chains and biological states, repair common structural problems, and remove irrelevant crystallographic contacts.
2. Collect interface evidence. Record known contact residues, cross-links, mutational effects, conservation, or predicted interaction sites. Distinguish direct restraints from lower-confidence hints.
3. Choose global or restrained search. Use global methods when the interface is unknown and information-driven methods when defensible restraints can reduce the search space.
4. Run the comparison workflow. Generate models with HADDOCK3, LightDock, EquiDock, and AF2Dock while preserving each method’s clusters, rankings, and native structures.
5. Compare candidate interfaces. Inspect recurring contacts, buried surface geometry, clashes, restraint satisfaction, and method agreement. Use DockQ only when a native reference complex is available.

How to interpret protein–protein docking results

Prioritize interface clusters supported by independent evidence rather than selecting a single model solely because it ranks first. Similar interfaces produced by unrelated methods can increase confidence, but correlated training data or scoring assumptions can also create shared bias.

A docked complex does not establish that the proteins interact in the cell, in the modeled stoichiometry, or under the relevant conditions. Validate contact residues and interaction-dependent function with suitable biochemical, biophysical, structural, or cellular experiments.

How the protein–protein docking workflow works

Run HADDOCK3, LightDock, EquiDock, and AF2Dock in parallel and retain each method’s models for comparison.

1. Prepare both partners. Upload separate PDB structures, select the correct chains and biological states, repair common structural problems, and remove irrelevant crystallographic contacts.
2. Map interface evidence. Record known contact residues, cross-links, mutational effects, conservation, or predicted interaction sites. Distinguish direct restraints from lower-confidence hints.
3. Choose global or restrained search. Use global methods when the interface is unknown and information-driven methods when defensible restraints can reduce the search space.
4. Generate complex models. Generate models with HADDOCK3, LightDock, EquiDock, and AF2Dock while preserving each method’s clusters, rankings, and native structures.
5. Compare interfaces. Inspect recurring contacts, buried surface geometry, clashes, restraint satisfaction, and method agreement. Use DockQ only when a native reference complex is available.

Inputs and outputs

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

Inputs

• Structural inputs. PDB SDF SMILES Two prepared PDB structures, optionally accompanied by residue-level interface restraints.
• Method context. Binding-site evidence, restraints, receptor-state provenance, known ligands, or reference complexes when available.

Outputs

• Docked structures. PDB PDBQT SDF Ranked complex models, method-native scores, interface contacts, and validation metrics when a reference is supplied.
• Review evidence. Method-native rankings, confidence, logs, interaction context, failures, and files for reproducible follow-up.