rewirebio.iobenchmarks
Dataset subset

POISK 1-in-50 screening targets (27 antigen constructs)

The 27 antigen constructs with at least 75 patent-derived binders each, used to build one-binder-in-fifty candidate pools for the screening simulation.

Evaluation results

8 evaluations · 14 results. Different protocols are not a single leaderboard.

Filter evaluations

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configurationProtocol and datasetFindingEvidence and details
Configuration: AlphaFold-Multimer with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.12 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

AlphaFold-Multimer on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "AlphaFold-Multimer", column "1-in-50", sub-column "Top-1 Acc."
Configuration: AlphaFold-Multimer with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
11.7% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 1.5. Table 8 also prints a standard deviation across targets for this engine (13.4 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

AlphaFold-Multimer on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "AF-M"
Configuration: Boltz-2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.03 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

Boltz-2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "Boltz-2", column "1-in-50", sub-column "Top-1 Acc."
Configuration: Boltz-2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
3.1% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 0.8. Table 8 also prints a standard deviation across targets for this engine (3.3 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

Boltz-2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "Boltz-2"
Configuration: IntelliFold v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.25 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

IntelliFold v2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "IntelliFold v2", column "1-in-50", sub-column "Top-1 Acc."
Configuration: IntelliFold v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
25.2% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 1.7. Table 8 also prints a standard deviation across targets for this engine (19.0 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

IntelliFold v2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "IF v2"
Configuration: IntelliFold v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-10 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
56% Top-10 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: Not yet extracted: The spread for top-10 retrieval is shown only as a Figure 5 error bar with no printed number.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

IntelliFold v2 on POISK 1-in-50 screening simulation, top-10 retrieval

poisk-20261115-protocol-screen-1in50-top10

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Section 3.2.3, sentence beginning "Protenix-v2 and IntelliFold v2 demonstrated comparable retrieval in broader rankings", value for IntelliFold v2
Configuration: OpenFold3 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.12 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

OpenFold3 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "OpenFold3", column "1-in-50", sub-column "Top-1 Acc."
Configuration: OpenFold3 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
11.6% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 1.6. Table 8 also prints a standard deviation across targets for this engine (11.4 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

OpenFold3 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "OF3"
Configuration: Protenix-v1 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.23 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

Protenix-v1 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "Protenix-v1", column "1-in-50", sub-column "Top-1 Acc."
Configuration: Protenix-v1 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
23.3% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 1.9. Table 8 also prints a standard deviation across targets for this engine (19.5 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

Protenix-v1 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "Prot-v1"
Configuration: Protenix-v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
0.29 Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
fraction · higher

Uncertainty: Not reported by the source: Table 1 prints no spread for this column; the spread is in Table 8 and Figure 5.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

Protenix-v2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 1, row "Protenix-v2", column "1-in-50", sub-column "Top-1 Acc."
Configuration: Protenix-v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-1 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
28.9% Top-1 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: SD 1.4. Table 8 also prints a standard deviation across targets for this engine (20.6 percentage points), which measures between-target spread rather than run-to-run spread and is kept as a separate claim record.

Coverage: 540/540 discovery trials per seed scored, 25 repeats

Author-reported evaluation · Source checked
Methods, coverage and source

Protenix-v2 on POISK 1-in-50 screening simulation, top-1 retrieval

poisk-20261115-protocol-screen-1in50-top1

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Table 8, row "Mean over targets", column "Prot-v2"
Configuration: Protenix-v2 with full-Fv ipSAE rankingProtocol: POISK 1-in-50 screening simulation, top-10 retrieval
Dataset subset: POISK 1-in-50 screening targets (27 antigen constructs)
54.7% Top-10 accuracy (retrieval; one patent-supported binder ranked against 49 putative decoys for the same antigen by full-Fv ipSAE, mean over 27 antigen constructs, 20 trials per construct per seed, 25 seeds)
percent · higher

Uncertainty: Not yet extracted: The spread for top-10 retrieval is shown only as a Figure 5 error bar with no printed number.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

Protenix-v2 on POISK 1-in-50 screening simulation, top-10 retrieval

poisk-20261115-protocol-screen-1in50-top10

Aggregation: Not reported

Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF) · Section 3.2.3, sentence beginning "Protenix-v2 and IntelliFold v2 demonstrated comparable retrieval in broader rankings", value for Protenix-v2

Source checking is not independent reproduction. Release 2026-11-21-c8bea25e5abb.

Research readiness

0 of 4 readiness checks met. These checks assess whether the evidence supports a reproducible investigation; a source-checked score alone does not meet them.

Readiness checks, gaps and artifacts

Release 2026-11-21-c8bea25e5abb · Evidence verified: Not verified

Evidence incomplete

Replay metrics

Exact outcomes, predictions, identifiers and evaluator are connected.

Missing or unresolved evidence

  • No verified artifact manifest is linked to this exact record.
  • File checksums match the recorded files: not yet verified
  • Predictions are matched to the right samples: not yet verified
  • Score meaning and direction are confirmed: not yet verified
  • Metrics are recomputed from the saved predictions: not yet verified

Verified: Not verified

Evidence incomplete

Investigate discrepancies

Replay evidence includes annotations and an assessment of dependence. Unknown independence permits descriptive analysis only.

Missing or unresolved evidence

  • No verified artifact manifest is linked to this exact record.
  • File checksums match the recorded files: not yet verified
  • Predictions are matched to the right samples: not yet verified
  • Score meaning and direction are confirmed: not yet verified
  • Metrics are recomputed from the saved predictions: not yet verified
  • Sample annotations are recorded: not yet verified
  • Dependence between samples is assessed: not yet verified

Verified: Not verified

Evidence incomplete

Run locally

A pinned recipe describes the inputs, environment and resource requirements.

Missing or unresolved evidence

  • No verified artifact manifest is linked to this exact record.
  • File checksums match the recorded files: not yet verified
  • Predictions are matched to the right samples: not yet verified
  • Score meaning and direction are confirmed: not yet verified
  • A pinned run recipe exists: not yet verified
  • Compute requirements are estimated: not yet verified

Verified: Not verified

Evidence incomplete

Validate independently

Separate data and exposure records support an independent test.

Missing or unresolved evidence

  • No verified artifact manifest is linked to this exact record.
  • File checksums match the recorded files: not yet verified
  • Predictions are matched to the right samples: not yet verified
  • Score meaning and direction are confirmed: not yet verified
  • Independent validation data exist: not yet verified
  • Overlap with training data is checked: not yet verified

Verified: Not verified

Readiness describes the evidence in this release. Availability on your computer is checked separately when an investigation runs. Existing data exposure can prevent independent validation even when files are available.

Artifacts and reproduction

No verified artifact manifest is connected to this record yet. The gaps above identify what is needed before analysis can begin.

Subset and evaluation context

This record describes a particular subset or cohort used in an evaluation. Its results do not describe the full dataset.

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

12 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance
Independent groups
27
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.independent_groups

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Limitations
2 values
  • Distinct constructs can share a UniProt accession, so 27 constructs are fewer than 27 distinct proteins: Q02223 and P18627 each appear twice in Table 8.
  • Restricted to targets with many patent-derived binders, which is not a random sample of screening problems.
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.limitations

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Source dataset
poisk-20261115-data-cofolding-benchmark-22494
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.source_dataset

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Source location
Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.source_locator

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Split
Evaluation only
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.split

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Subset scope
27 targets from the benchmark that possessed at least 75 patent-derived binders. Each trial draws one binder and 49 putative non-binders for the same antigen; 20 trials per target give 540 discovery trials, repeated under 25 scenario-generation seeds.
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.subset_scope

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Test count
540
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.test_count

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Version
As described in bioRxiv v1 Section 3.2.3
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.version

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Description
The 27 antigen constructs with at least 75 patent-derived binders each, used to build one-binder-in-fifty candidate pools for the screening simulation.
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: description

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

Relationship: same data as
poisk-20261115-data-cofolding-benchmark-22494
Context-only references
Tupikina et al. 2026, Can Folding Models Tell Binders from Bluffers? Evidence from POISK (bioRxiv v1 PDF)

Original source ↗

Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Version: v1, posted 26 October 2026
Retrieved: 2026-11-15T23:32:24Z

not individually reviewed

No individual claim review recorded

Audit details

Field: links:same_data_as:poisk-20261115-data-cofolding-benchmark-22494

Source artifact SHA-256: 526def150fb5ae8f9ac4533efcf16c86e547e636cdb9bfba9a574cd1e2a2f924

Hash scope: One download of 2026-11-15. The PDF container is not byte-stable: bioRxiv regenerates it on request, so artifact_sha256 is that one download's hash and no refetch reproduces it. Three downloads on 2026-11-16 gave 3389c1bf6d3654c5b2151a07fcfcefc67cf6f36e5f0facfc03c33521d8b84491 twice and d699fd680e3b0097ca80130c288e8d872773dc1e523044f4605e06fc4f89d9bc once. All three yield the same pdftotext -layout text layer, SHA-256 e75c18130ca43770d2f0ed05ae96528759666c0e4fa08f7725f5a32b21187696, which is the article's stable identity, is archived gzipped in artifacts/biorxiv-v1-layout.txt.gz under archive_sha256, is what extract/build_batch.py asserts, and is what every stored value was read from.

Format: application/pdf, 16 pages

How to read the evidence

Sources and history

Release 2026-11-21-c8bea25e5abb · Record review: source checked

1 source record and release historyDownload this release (gzip)
Technical metadata and record identifiers

Technical metadata for this record is tracked under stable ID poisk-20261115-data-screening-27-targets.

Source locator: Section 3.2.3 '1-in-50 zero-shot screening simulation'; Table 8 columns '#' and 'UniProt'

Related records

Suggest a correction