ABOUT PLATFORM PRICING TEAM BLOG

What is the DockQ score? And when to trust it.

A practical guide to ranking docking models and reading the score.

Download

Summary

DockQ turns three interface signals into one score so you can sort docking models fast. Fnat asks if the right residues are touching, iRMSD measures how much the interface backbone moved, and LRMSD checks how far the ligand moved after superposing on the receptor. The score runs from 0 to 1 and most groups triage with these bands: 0.80 or higher high, 0.49 to 0.79 medium, 0.23 to 0.48 acceptable, less than 0.23 poor.

DockQ is now a standard validation readout in docking and structure prediction, used in many papers to benchmark, reproduce results, and cross compare methods. Trust DockQ most when there is modest conformational change and a good reference, and add extra checks for induced fit, symmetry, or low confidence references.

How to use DockQ on Vici.bio

A. Upload a predicted complex, the model structure, and a trusted reference, the native structure.

B. Click Execute to get DockQ, Fnat, iRMSD, LRMSD, plus interface and clash checks.

C. Sort by DockQ, then skim Fnat and the contact map to confirm the right residues touch.

Run DockQ on Solo or batch score in NEXO . DockQ reads PDB and mmCIF, handles multiple interfaces, nucleic acids, and small molecules, and performs automatic chain mapping for symmetric or multimeric assemblies.

Mini case study

Pose A

GO
DockQ 0.82
Fnat0.76
iRMSD1.4 Å
LRMSD2.1 Å
Fnonnat0.08
F10.76

Contacts correct, small movements. Send to refinement & wet lab.

Pose B

KEEP
DockQ 0.56
Fnat0.68
iRMSD2.6 Å
LRMSD6.5 Å
Fnonnat0.22
F10.66

Mostly right contacts; consider loop relaxation or local redocking.

Pose C

DROP
DockQ 0.18
Fnat0.20
iRMSD5.8 Å
LRMSD14.0 Å
Fnonnat0.55
F10.25

Off-target contacts dominate; large displacements.

Use cases

Antibody screening

Rank paratopes or designs before wet lab.

  • Sort by DockQ, confirm Fnat with contact map.
  • Keep medium DockQ if CDRs are flexible.

Enzyme inhibitor poses

Quick pass before FEP or MD.

  • Use LRMSD + clash flags to kill bad poses.
  • Cross-check key contacts along the pocket.

Protein–Nucleic Acid modeling

Verify TF / RNA interfaces.

  • iRMSD uses appropriate backbone atoms.
  • Prefer sequence-specific contacts recovered.

Symmetry & homomers

Validate labels & mapping.

  • Use automatic chain mapping.
  • Read per-interface scores and GlobalDockQ.

Benchmarking protocols

Compare methods with one metric.

  • Report DockQ, Fnat, iRMSD, LRMSD.
  • State clash / mapping checks in methods.

QC for datasets

Curate before training.

  • Detect clashes, numbering, missing ligands.
  • Gate designs by DockQ band.

Native structure selection guide

  • Use the best experimental complex available. Note engineered mutations near the interface.
  • If only monomers exist, curate a high-confidence complex as a stand-in native and document the build (templates/constraints/docking).
  • Cross-check the reference: interface visualization, chain IDs, biological assembly vs. crystal contacts, and reasonable ligand pose.
  • When no experiment exists, build a consensus from ipTM, interface pAE, and conservation; treat DockQ bands as guidance, not hard gates.

File format tips

  • Prefer mmCIF for large assemblies and consistent residue numbering; PDB is fine for small complexes but watch chain IDs & insertion codes.
  • Keep chain labels stable between model and native. If labels differ, rely on automatic chain mapping and review the mapping report.
  • Clean inputs before scoring: remove unwanted alt locs, resolve missing atoms near the interface, and ensure ligands of interest exist in both files.
  • For small molecules, align protonation & tautomer state across model and native to keep LRMSD meaningful.

Comparing DockQ to other signals

  • plDDT shows local backbone confidence; high plDDT with low DockQ often means a confident but misplaced pose.
  • ipTM tracks overall assembly confidence. Agreements with DockQ are strong wins; disagreements flag docking register issues.
  • Interface pAE highlights interface uncertainty. Low pAE with high DockQ is ideal; high pAE with medium DockQ suggests instability.
  • Estimated ΔG / scoring adds physics. Favorable ΔG with poor DockQ often indicates scoring false positives—seek agreement.

Most reliable when

  • Partners bind without major remodeling (interfaces are rigid or near-rigid).
  • You compare against a high-quality experimental reference.
  • Targets are standard protein–protein interactions without tricky symmetry.

Use caution & add checks

  • Induced fit / flexible loops (e.g., antibody CDRs): RMSD penalties can look harsh even when key contacts are right; inspect Fnat and the contact map.
  • Symmetry / multimers: wrong chain assignments distort scores; use automatic chain mapping and check GlobalDockQ.
  • Low-confidence references: be conservative; compare replicates and look for consistent trends across Fnat, iRMSD, LRMSD.
  • Tiny interfaces / sparse contacts: Fnat is noisy; small changes can swing the score.
  • Clashes or labeling errors: clashes and misnumbered chains inflate/deflate components; run clash checks and do a quick visual sanity pass.

Reporting templates, figures and tables

Per-target compact table

Model ID DockQ Fnat iRMSD (Å) LRMSD (Å) Fnonnat F1 Notes
A_17 0.84 0.78 1.3 2.0 0.06 0.77 mapping verified
B_02 0.58 0.69 2.7 6.8 0.20 0.66 loop relax recommended
C_41 0.19 0.22 5.6 13.9 0.52 0.27 clash flagged

Mini figure previews

DockQ bars
DockQ vs LRMSD
Contact map

How to present results.

(1) Per-interface bars: sort interfaces by DockQ (high → low) and annotate only outliers. Color bars by band (acceptable/medium/high).
(2) DockQ vs LRMSD: plot DockQ on y, LRMSD on x with a light grid hatch; upper-left is best. Label just a few representative models to avoid clutter.
(3) Contact map panel: show native, model, and overlap mini-maps. Use the same binning across targets and include a legend once.
Docs: keep axis ranges consistent across figures, state the DockQ version and preprocessing (mapping/clash checks), and put quick actions (e.g., “loop relax recommended”) in the table’s Notes column to accelerate triage.

How it works

DockQ answers two questions. First, are the correct interfacial contacts present, which is Fnat. Second, how much did the interface move, which is iRMSD for interface backbone atoms and LRMSD for the ligand after superposing on the receptor. DockQ blends these into one number by rescaling the RMSDs so large displacements do not overwhelm contact recovery. The exact equation appears in the figure. Use the sliders under the figure to change Fnat, iRMSD, and LRMSD, and watch the DockQ score update in real time so the tradeoffs are obvious.

DockQ= 13 [ Fnat + 1 1+ (iRMSD1.5) 2 + 1 1+ (LRMSD8.5) 2 ]

Play with DockQ

Adjust the components to see the DockQ score update.

0.60 2.5 Å 10.0 Å
0.53

Frequently asked Questions

References

  1. Basu, S. and Wallner, B. (2016) DockQ, a quality measure for protein protein docking models. PLOS ONE 11(8):e0161879.
  2. Mirabello, C. and Wallner, B. (2024) DockQ, improved automatic quality measure for protein multimers, nucleic acids, and small molecules. Bioinformatics 40(10):btae586.
  3. Desta, I. T. et al. (2020) Performance and its limits in rigid body protein protein docking. Structure 28(9):1071–1081.e3.
  4. Collins, K. W. et al. (2024) The CAPRI resource for assessment of modeling protein interactions. Journal of Molecular Biology 436(14):168446.