AI for Bio · Benchmark 28

Adaptyv competitions / BenchBB

Multi-round prospective protein design evaluation with public data; run by a commercial service provider, not a board to optimise against.

bench-28Type A+BEvidence L5Proteins● Active No ranking中文

Overview

NameAdaptyv competitions / BenchBB · aliases: BenchBB, Adaptyv protein design competition
Full nameAdaptyv protein design competition series and the BenchBB target set
Ambiguous names (not auto-linked)None
FamilySeveral rounds from one organiser plus a static target set; targets and definitions differ, so hit rates do not compare across rounds.
  • Competition round 1 · †
  • Competition round 2 · †
  • BenchBB ·
  • 2025 competition round ·
  • 2026 joint competition ·
DomainProteins
InclusionOriginal 30
TypeA B · Periodic prospective wet-lab competition plus a target set
Evidence levelL5 · Prospective wet lab; no independent replication
PublisherAdaptyv Bio (Platform company)
First release† unverified
Basis: Organiser post announcing the first competition round; the day was inferred from a post title, only the month is supported source
Other milestones
  • Preprint v1: · BenchBB community paper, bioRxiv v1 source
  • Challenge opens: † · First competition round announced (month only) source
Latest update · new round
Basis: Organiser competition page: final challenge window of the 2026 joint competition opened
Challenge 5 of the 2026 joint competition opened on 2026-10-26; five weekly challenges all closed to submissions by 2026-11-09. source
The most recent sourced event, not a claim that nothing changed since.
Next expected† partly verified · results
Basis: Organiser schedule: lab data to be published in early January 2027 (schedule marked subject to change)
Note: The same page also states testing completes 2026-12-07; the site's earlier 2026-12-15 has no counterpart source
Status● Active · Active with a new round each quarter; results of the 2026 joint competition pending
ScaleRound 1: 1,857 submissions, about 600 tested; 2025 round: 1,196 tested; 2026 joint run: 1,320 sequences across 16 targets; BenchBB: 7 targets.
  • Round 1 submissions / tested: 1,857 / ~600 (Round 1)
  • 2025 round sequences tested: 1,196 sequences (2025 round)
  • 2026 joint run sequences / targets: 1,320 / 16 (2026 joint)
  • BenchBB targets: 7 targets (BenchBB)
Wet-lab refreshFresh experiments each round · One round per quarter, about 300-1,200 sequences per round, measured in the organiser automated lab.
Own wet lab needed to evaluate a new modelYes, to claim real generalization
Models reported4 counted (board 2, modified protocol 1, retrospective 1)
Unverified fields† first_release, family.members[0], family.members[1], next_expected, models_reported[0]
As of2026-11-16

What it measures

  • Task: Whether submitted designed sequences express and bind the stated target.
  • Truth: Binding measurements from the organiser lab under one protocol, including measured negatives.
  • Primary metric: Binding hit rate
  • Secondary metrics: Affinity ranking; Expression rate
  • Version traps: Targets, hit thresholds and denominators differ by round; tested sets are not random samples but pre-screened by computational metrics.

Current ranking

No ranking

MetricBinding hit rate and affinity ranking, plus expression rate
Eval setAnthropic x Adaptyv 2026 competition, Challenges 1-5 (EGFR, TNF-a, GLP1R, TfR1, KRAS G12D pMHC)
Snapshot
Boardhttps://proteinbase.com/competitions/anthropic-adaptyv-2026
Retrieved2026-11-16

Caveat. The current round has no results: all five challenges are closed, the organiser page shows no board, ranks or scores, and data are scheduled for early 2027; past results sit in organiser posts with differing definitions.

No headline entry: The current round has no results and the organiser page shows no ranks or scores.

No ranking entries recorded.

Models reported

Grouped by relation type; each line shows its source type. Counted relations add up to 4. A pairing whose evidence rests only on this survey's own page is shown as unverified.

Official leaderboard or results (2)

  • BindCraft Board Nipah and EGFR rounds · The developers took part; an early-round top-10 placement comes from an organiser post (second-hand) †partly verified · evidence: link
  • Mosaic Board · Entrant in the 2025 de novo track with results on the organiser site; hit rates are self-reported · evidence: link

Developer report, modified protocol (1)

  • Proteina-Complexa Self-reported · Added by R7/fc_zero as a missing candidate and confirmed in the review. Resolution: The project page states participation and links the organiser's competition page, but the organiser page does not name the model, so this is self-reported participation that the organiser has not confirmed - which the schema routes to self-reported-modified-protocol, not official-participation. Established, so this model also leaves the zero-pairing list. · evidence: link

Retrospective dataset use (1)

Caveats

  1. One target and a few hundred designs per round; the organiser is a commercial service provider; computational proxies correlate only weakly with measured binding; rounds are not comparable.
  2. Tested sets are pre-screened by computational metrics, so correlations are computed on a truncated range and understate the true value.
  3. Rounds mix redesign of known antibodies with de novo design; these differ in difficulty and the former is a strong baseline.
  4. Replication requires paying for measurement, there is no independent lab replication, and the organiser has a commercial interest.
  5. BenchBB has no common scoring protocol and its board does not publish submission counts.

How to use it

It cannot be run statically: there is no reusable train/test split, so designs must be sent for measurement or entered in the next round. Report hit threshold, denominator, expression rate and target, add orthogonal validation; a single round cannot rank methods or speak to function, specificity or developability.

Pairs well with: AIntibody · FLAb / FLAb2 · AbBiBench

Sources

  1. Organiser page for the 2026 joint competition (official site)
  2. Round 2 results (organiser post) (company blog)
  3. BenchBB and community introduction (organiser post) (company blog)
  4. BenchBB community paper (bioRxiv) (preprint)
  5. Meta-analysis of computational proxies (bioRxiv 2025) (third-party paper)
  6. Organiser post archive (company blog)

Corrections

4 corrections
FieldWasNowBasisFound
members[{"name": "EGFR round 1", "first_release": "2024-07-18", "basis": "announcement post"}, {"name": "EGFR round 2", "first_release": "2024-10", "basis": "launched Oct 2024, results Dec 2024"}, {"name": "Anthropic x Adaptyv 2026", "first_release": "2026-09-28", "basis": "Proteinbase competition page"}][{"name": "Binder design competition round 1 (EGFR)", "first_release": "2024-07", "basis": "announced July 2024; exact day inferred from a post title only, flagged"}, {"name": "Round 2 (EGFR)", "first_release": "2024-11", "basis": "existing note, day not re-verified"}, {"name": "BenchBB / community paper", "first_release": "2025-04", "basis": "bioRxiv 2025.04.17.648362"}, {"name": "Nipah protein design competition", "first_release": "2025-10-21", "basis": "announced 2025-10-21; submissions open 2025-11-01; leaderboard 2025-11-19; results 2026-01-05"}, {"name": "Anthropic x Adaptyv 2026", "first_release": "2026-10-26", "basis": "challenge 5 window opened 2026-10-26 (organiser page)"}]source2026-11-16
wetlab_noteabout $149 per sequencenot stated by the operator; the figure appears only in secondary sources and no pricing page was locatedsource2026-11-16
next_expectedresults published 2026-12-15organiser schedule: testing by 2026-12-07, data published early January 2027 (subject to change)source2026-11-16
models_reported.model-proteina-complexanot listed (zero-pairing list)self-reported-modified-protocolsource2026-11-16

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