AI for Bio · Benchmark 22

CAPRI

The oldest prospective blind test for protein complexes; its classes and DockQ are the common protocol

bench-22Type AEvidence L4Structure & co-folding◔ Round in progress No ranking中文

Overview

NameCAPRI · aliases: CAPRI, CAPRI Round 57, CAPRI Round 65, CASP16-CAPRI, CASP15-CAPRI
Full nameCritical Assessment of PRedicted Interactions
Ambiguous names (not auto-linked)DockQ
FamilyNot a family entry. Successive rounds of one challenge rather than a family entry; joint rounds with CASP share targets but are assessed separately.
DomainStructure & co-folding (also: Proteins)
InclusionOriginal 30
TypeA · Periodic blind test with a scorer track
Evidence levelL4 · Truth unpublished at submission; no wet-lab functional validation
PublisherCAPRI management group, EMBL-EBI Protein Data Bank in Europe (Mixed)
First release† partly verified
Basis: The official site confirms only the year of the first round; the month is unverifiable, with a secondary paper saying July 2001 and the design meeting held in June 2001 source
Other milestones
  • First public evidence: † · The official round list gives no dates for the early rounds source
  • Challenge opens: † · Year of the first round, per the official site source
Latest update · new round
Basis: Official round page: Round 66 opened for predictions, with registration from 2026-10-20 to 11-19
Round 66 opened for predictions, with registration to 2026-11-19, submissions closing 11-20 and scoring on 11-23 to 11-27; three nanobody targets, co-organized with a 2026 meeting. source
The most recent sourced event, not a claim that nothing changed since.
Next expected · new round
Basis: Start of the Round 66 scoring window as published on the official round page source
Status◔ Round in progress · Round in progress: Round 65 is closed pending publication and Round 66 is open for predictions
ScaleOver 60 rounds and 160+ targets cumulatively; Round 57 (CASP16-CAPRI) had 34 targets and Round 65 has 54.
  • Rounds cumulatively: 60+
  • Targets cumulatively, per a 2020 organizer statement: 160+
  • Targets: 37 (Round 54 (CASP15-CAPRI))
  • Models submitted: 21941 (Round 54 (CASP15-CAPRI))
  • Targets: 34 (Round 57 (CASP16-CAPRI))
  • Targets: 54 (Round 65 (CASP17-CAPRI))
Wet-lab refreshFresh experiments each round · About one joint round every two years with 30-40 targets; truth comes from experimental structures supplied before publication, not measured for the benchmark.
Own wet lab needed to evaluate a new modelNo, run it as a download
Models reported2 counted (board 2, unconfirmed 1)
Unverified fields† first_release, release_milestones.challenge_open, release_milestones.first_public_evidence, ranking.entries[2], scale.items[1]
As of2026-11-16

What it measures

  • Task: Two tracks: predict complex structures from component sequences, or pick near-native models out of the pool of all submissions in the scorer track.
  • Truth: X-ray or cryo-EM structures, supplied by structural biologists and unpublished at prediction time.
  • Primary metric: The four CAPRI quality classes from Fnat, L-RMSD and i-RMSD
  • Secondary metrics: DockQ, a continuous 0-1 score; Per-round counts of targets by quality class
  • Version traps: State the round, target class and number of submitted models: the protocol allows external MSAs, templates and unlimited compute, so rankings mix in sampling effort.

Current ranking

No ranking

MetricCAPRI quality classes and DockQ
Eval setRound 65 (CASP17-CAPRI) closed in Aug 2026 with results unpublished; Round 66 open with scoring 2026-11-23 to 11-27; entries come from Round 57 (CASP16-CAPRI, 2024)
Snapshot stale
Boardhttps://www.ebi.ac.uk/pdbe/complex-pred/capri/round/65/
Retrieved2026-11-16

Caveat. Round 65 results are to appear in the CASP17 special issue and no ranking exists yet; CAPRI publishes per-target class counts rather than one board.

No headline entry: There is no ranking this round, and the entries come from the CASP16 assessment of the joint round rather than a CAPRI-own standing.

RankModelScore (verbatim)Metric · eval set / protocolSource type · dateSource and note
10AF3-serverrank 10 of 81 (CASP16 protein multimers); avg DockQ 0.53 first-rankedCAPRI quality classes and DockQ
Round 65 (CASP17-CAPRI) closed in Aug 2026 with results unpublished; Round 66 open with scoring 2026-11-23 to 11-27; entries come from Round 57 (CASP16-CAPRI, 2024)
Third-party
source From the CASP16 oligomer ranking assessed jointly with CAPRI, not a CAPRI-own ranking.
30MassiveFoldrank 30 of 81 (CASP16 protein multimers)CAPRI quality classes and DockQ
Round 65 (CASP17-CAPRI) closed in Aug 2026 with results unpublished; Round 66 open with scoring 2026-11-23 to 11-27; entries come from Round 57 (CASP16-CAPRI, 2024)
Third-party
source Same third-party assessment table; the MassiveFold paper on CASP16-CAPRI (31 targets, 65 interfaces) gives no rank.
—Top-performing academic group (CASP16 oligomers)Withheld: The figure appears only in a secondary summary and was not re-verified in primary material, so it is not shown.CAPRI quality classes and DockQ
Round 65 (CASP17-CAPRI) closed in Aug 2026 with results unpublished; Round 66 open with scoring 2026-11-23 to 11-27; entries come from Round 57 (CASP16-CAPRI, 2024)
Third-party
†
source second-hand Not a site model; the official assessment calls complex prediction unsolved, with over 30% of targets, especially antibody-antigen, highly difficult.

Models reported

Grouped by relation type; each line shows its source type. Counted relations add up to 2. A pairing whose evidence rests only on this survey's own page is shown as unverified.

Official leaderboard or results (2)

  • AF2-Multimer Board CASP15-CAPRI official assessment, assessor-run yardstick baseline · The official round assessment lists it as the assessors' yardstick baseline, counted as an official entry; no per-target numbers. †partly verified · evidence: link
  • MassiveFold Board CASP16-CAPRI Round 57 (Round 58 data also released) · Took part in Round 57 (CASP16-CAPRI); self-reports 31 targets and 65 interfaces, with no rank. †partly verified · evidence: link

Unconfirmed (1, not counted)

  • AlphaFold 3 Unconfirmed source CASP16 complexes · No standalone entry appears in the official CAPRI assessment material; only secondary sources call it a component of most pipelines. †disputed · evidence: link

Caveats

  1. Each round has only tens of targets with an uncontrolled class mix, so a single target can dominate aggregate scores and confidence intervals are usually absent.
  2. It shares targets and submissions with the CASP complex track, so it is not independent evidence.
  3. The protocol allows external MSAs, templates and unlimited compute, so rankings mix in sampling and compute, and there is no common baseline suite.
  4. Easy homo- and heterodimers are near saturation while antibody-antigen, flexible complexes and large assemblies remain unsolved.
  5. No study links the scores to affinity, specificity or design hit rates, and targets and submissions are scattered with no versioned dataset.

How to use it

A real blind test means entering the next round. Historical targets can be replayed statically: take joint-round targets released after the training cutoff, submit 5 models, score with DockQ, report per target rather than only a mean, and keep antibody-antigen separate. Models trained on the full PDB risk leakage on old targets; for a de-leaked static complement see the PLINDER family.

Pairs well with: CASP · PLINDER / PINDER / Runs N' Poses · FoldBench · CAMEO

Sources

  1. Official CAPRI site with the round list (official site)
  2. Round 65 page (official site)
  3. Assessment paper for Round 54, CASP15-CAPRI (journal)
  4. Third-party assessment paper on CASP16 structure prediction (third-party paper) ·
  5. DockQ paper (journal)
  6. DockQ v2 (preprint)
  7. Review on the origins of CAPRI (journal)

Corrections

7 corrections
FieldWasNowBasisFound
first_release2001-072001source2026-11-16
first_release_basisRound 1 began soon after Charleston meeting (2001-06-16..19) per CAPRI site; a paper says July 2001; EBI round page shows odd 2000-11-29 registration, unreconciledfirst CAPRI round, 2001 (year-only evidence on the official site). Month unverifiable: a secondary paper says July 2001, the Charleston design meeting was 2001-06-16..19, and the EBI round-1 page shows placeholder 2000 dates.source2026-11-16
first_release2001-062001-07source2026-11-16
latest_updateRound 57 (CASP16-CAPRI, 2024) described as the latest round2026-11-02: Round 66 opened for predictions; Round 65 (54 targets) closed in Aug 2026 with results unpublishedsource2026-11-16
models_reportedAF2-Multimer listed as having no CAPRI pairingofficial-participation: the assessors' baseline in the official round assessmentsource2026-11-16
ranking.entriesa top group's success rate above 60% shown as a numberwithheld: the figure appears only in a secondary summarysource2026-11-16
statusactiveround-in-progress (Round 66 predictions open, scoring 2026-11-23 to 11-27)source2026-11-16

Part of the AI for Bio benchmarks survey · all 44 benchmarks · open in the explorer. Record as of 2026-11-16.