HADDOCK3

What is HADDOCK3?

HADDOCK3 is a molecular docking platform for building protein complexes. It can use known or predicted interface residues to guide docking, which is useful for antibody-antigen, enzyme-partner, and other protein-protein systems.

The tool supports two approaches. Data-driven docking uses interface information supplied as residues or restraint files. Ab initio docking searches without known interface residues. In ProteinIQ, the second uploaded protein is called the ligand; it is not a small molecule.

How to use HADDOCK3 online

Upload a receptor and a second protein in PDB format. Add interface residues when binding-site information is available, or leave them empty for ab initio docking. ProteinIQ runs docking, refinement, clustering, and scoring, then returns ranked complex structures, scores, and the complete run files.

Choose the docking mode

There is no separate data-driven mode switch. The inputs determine the mode automatically.

For data-driven docking, open Advanced settings and fill in the receptor and ligand interface residue fields. Enter active residues and, optionally, passive residues as comma-separated numbers or ranges, such as 38,40,45-50. Add a chain ID for a specific chain. Residues must be defined for both partners, with at least one active residue across the pair.

The Ab initio mode control stays visible when interface restraints are entered, but it does not affect a data-driven run.

Inputs

Active and passive residue fields generate ambiguous interaction restraints. An uploaded ambiguous-restraint file is an alternative to those fields. Exact evidence such as cross-links belongs in an unambiguous .tbl file.

Either docking partner can contain a PDB ensemble with MODEL and ENDMDL records. All models are retained for native docking. The optional reference complex is a single structure.

Settings

The defaults are suitable for a standard run.

Increasing the initial model count or the pre-refinement selection broadens the search but increases runtime. Advanced settings are best left unchanged unless a run has a specific sampling or clustering problem.

Results

Each result row represents one complete receptor-ligand complex. Lower HADDOCK scores rank better within the same run.

Top models to return is a limit, not a promise. A request for 10 can return fewer models when:

• only a small number of refined models form a cluster;
• clusters contain fewer models than requested;
• some models fail because of clashes, topology problems, or incompatible restraints.

The Files tab contains every final ranked complex, the effective configuration, the chain and residue mapping, and two archives. haddock3_results.tgz contains final models, native evaluation and clustering data, analysis reports, and the native log. haddock3_reproducibility_bundle.tgz contains the submitted and prepared structures, restraints, configuration, and mapping. The separate run.log gives a concise run summary. In the native archive, clustfcc.tsv shows clustered and unclustered models, and capri_ss.tsv lists the models that reached final evaluation.

How HADDOCK3 works

Active residues are expected to participate in the interface. Passive residues may participate but are not penalized when they do not. HADDOCK3 converts these selections into ambiguous interaction restraints, so the run can satisfy plausible contacts without requiring one exact residue pair.

The workflow has four main parts:

• rigidbody generates and scores initial orientations.
• flexref allows local movement around the interface.
• emref minimizes the selected complexes.
• clustfcc groups similar interfaces before final model selection and CAPRI evaluation.

The final score combines interaction energies and restraint violations. It is most useful for ranking models from the same protein pair. Scores from unrelated systems should not be compared as if they were binding affinities.

Interpreting a run

A convincing result usually combines a favorable score, low restraint-violation energy, and several similar models in the same cluster. A single highly ranked model without cluster support is weaker evidence that docking converged on that binding mode.

Fragmented clusters can indicate an underdefined interface. More sampling or reliable interface restraints may help. Very restrictive or incorrect restraints can instead force poor contacts, so restraint choices should be checked when air energy remains high.

CAPRI metrics require a meaningful structural reference. Without an uploaded reference, HADDOCK3 compares models with the lowest-scoring model from the run; those values describe similarity within the run, not accuracy against an experimentally known complex.

When to use HADDOCK3

HADDOCK3 is most useful when experiments, literature, or predictions identify part of a protein-protein interface. LightDock is useful for broader exploration when no interface information is available.

For docking a small molecule into a protein binding site, use AutoDock Vina, GNINA, or DiffDock. This HADDOCK3 workflow requires protein atoms in both docking partners.

Large conformational changes are outside the main scope of this workflow. Local interface flexibility is modeled, but domain rearrangements and extensive disorder can limit accuracy.

Multi-chain uploads are supported as one docking body per uploaded partner. ProteinIQ preserves the original subunits with restraints and includes the chain and residue mapping in the result files. When interface residues are used, insertion-coded residue numbers must be normalized before submission.

Troubleshooting

PDB Fixer can help prepare structures with missing atoms or malformed records before docking.