Extract experimental hit rates of de novo designed protein binders from 27 open-access papers

Produce one CSV that records, for every design campaign reported in 27 listed open-access papers (2024–2026), how many computationally designed binder candidates were tested in the laboratory, how many of them bound their target, the criterion the paper used to call a design a binder, and the best reported affinity — each row citable to the exact place in the paper that states it.

Deep-learning binder design (RFdiffusion, BindCraft, AlphaProteo, ProteinMPNN pipelines and related methods) is usually reported as a "success rate" or "hit rate", but every paper counts differently: some count designs ordered, some designs expressed, some designs that passed a display screen. The numbers sit in figure captions, supplementary tables and single sentences of the main text, so no database holds them. This bounty buys a structured table of what these 27 papers report, with the criterion each paper applied carried alongside the numbers.

What a design campaign is

A design campaign is one combination of (paper, target, design method) for which the paper reports both:

• the number of distinct designed candidate binders that were experimentally tested for binding against that target, and
• the number of those candidates the paper reports as binding.

One row of the CSV describes exactly one campaign. A paper reporting designs against three targets with one method yields three rows. A paper reporting two different design methods against one target yields two rows. A paper reporting one method against one target in two sequential rounds (for example an initial design round and an affinity-maturation round) yields one row per round, and the round is named in design_method.

A campaign qualifies only if the binder candidates were produced computationally and are new sequences rather than a natural binder, a known antibody, or a variant panel of an already-published binder. Point-mutant or affinity-maturation panels built on a binder first reported in the same paper do qualify as a separate round. Panels built on a binder published elsewhere do not.

What counts as tested and as a binder

• n_designs_tested is the number of distinct designed sequences entering the experimental binding measurement the paper reports for that campaign. Where a paper reports a funnel (designs generated → ordered → expressed → assayed), use the number assayed for binding. Where a pooled library screen is the binding measurement, use the library size the paper states for that target.
• n_binders is the number of those designs the paper counts as binding its intended target, under the paper's own criterion.
• binder_threshold_as_stated is that criterion, quoted from the paper or paraphrased without changing any number, unit or assay, for example KD < 1 uM by SPR, enrichment in yeast display sort 2, detectable BLI response at 1 uM. If the paper states no criterion, write not stated.
• Numbers reported only as a percentage are converted only when the paper also states the denominator; otherwise the campaign is not extractable and the paper is handled under the provenance rule below.
• In-silico-only success rates (designs passing AlphaFold or Rosetta filters, with no laboratory measurement) are never a campaign.

Affinity and units

best_affinity_nM is the strongest (numerically lowest) dissociation constant the paper reports for any design of that campaign, in nanomolar, as a decimal number:

• pM → divide by 1000; µM (uM, μM) → multiply by 1000; mM → multiply by 1,000,000.
• Use the reported dissociation constants (KD). Only when the paper reports no KD for any design of that campaign, use the lowest reported apparent KD, EC50 or IC50 for that campaign instead.
• A value given with an error (45 ± 4 nM) is recorded as its central value: 45. A value given as a range (100–454 nM) is recorded as its lowest value: 100.
• Where a value is reported only as a bound (< 1 nM), record the bound's number: 1.
• If no affinity is reported for any design of that campaign, leave best_affinity_nM empty. An empty field is never wrong when the paper reports no value; it is wrong when the paper reports one.

Where a value may come from

Tables, figures, figure captions, main text and supplementary material of the listed open-access version are all acceptable sources, provided the row cites the location and quotes text from it that carries at least one of the row's numbers.

Schema, in this column order:

paper_doi,target_name,design_method,n_designs_tested,n_binders,binder_threshold_as_stated,best_affinity_nM,assay_type,table_or_figure,source_quote

• paper_doi is the DOI exactly as printed in sources.csv.
• target_name is the target as the paper names it, for example PD-L1, GDF15, SARS-CoV-2 RBD. Where the paper names a target and a site, name the target only.
• design_method is the method as the paper names it, plus the round where a paper reports more than one, for example RFdiffusion, BindCraft, RFdiffusion + ProteinMPNN (round 2).
• n_designs_tested and n_binders are non-negative integers, with n_binders at most n_designs_tested.
• best_affinity_nM is a decimal number in nM, or empty.
• assay_type is one of SPR, BLI, ITC, yeast display, phage display, flow cytometry, ELISA, pulldown, cell assay, or other: <name as the paper calls it>. Where several assays are used for one campaign, name the one the binder criterion is based on.
• table_or_figure identifies the location of the value, as printed: Table 2, Fig. 3c, Supplementary Table 1, Extended Data Fig. 5b, or main text.
• source_quote is a verbatim quotation of at most 300 characters from that location (a table cell or row, caption fragment or sentence) that states at least one of the row's n_designs_tested, n_binders or best_affinity_nM values. Whitespace and line breaks may differ from the paper. The other numbers of the row may come from elsewhere in the same paper.

sources.csv

paper_doi,pmcid,year,journal,title,open_access_url
10.1038/s41586-025-09429-6,PMC12507698,2025,Nature,One-shot design of functional protein binders with BindCraft,https://pmc.ncbi.nlm.nih.gov/articles/PMC12507698/
10.1038/s41467-025-67866-3,PMC12852815,2026,Nature Communications,Improved protein binder design using β-pairing targeted RFdiffusion,https://pmc.ncbi.nlm.nih.gov/articles/PMC12852815/
10.1002/advs.202502015,PMC12412469,2025,Advanced Science,A Complementarity‐Based Approach to De Novo Binder Design,https://pmc.ncbi.nlm.nih.gov/articles/PMC12412469/
10.1016/j.mtbio.2026.103336,PMC13267575,2026,Materials Today Bio,De novo design of dual-target mini-binders simultaneously neutralizing Streptococcus equi subspecies zooepidemicus M-like protein and host TNFR1 confers complete protection against lethal infection,https://pmc.ncbi.nlm.nih.gov/articles/PMC13267575/
10.1038/s12276-026-01727-x,PMC13234300,2026,Experimental & Molecular Medicine,De novo and scaffold-based design of GDF15 binders for cancer cachexia diagnostics and therapeutics,https://pmc.ncbi.nlm.nih.gov/articles/PMC13234300/
10.3389/fbioe.2026.1946220,PMC13736603,2026,Frontiers in Bioengineering and Biotechnology,Design and validation of nucleosome binding peptides using BindCraft,https://pmc.ncbi.nlm.nih.gov/articles/PMC13736603/
10.1021/acschembio.5c00774,PMC12723675,2025,ACS Chemical Biology,Evaluating BindCraft for Generative Design of High-Affinity Peptides,https://pmc.ncbi.nlm.nih.gov/articles/PMC12723675/
10.1073/pnas.2506269123,PMC12867733,2026,Proceedings of the National Academy of Sciences,Mapping targetable sites on the human surfaceome for the design of novel binders,https://pmc.ncbi.nlm.nih.gov/articles/PMC12867733/
10.1038/s41586-025-09721-5,PMC12727541,2025,Nature,Atomically accurate de novo design of antibodies with RFdiffusion,https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/
10.1002/advs.76202,PMC13339307,2026,Advanced Science,De Novo Designed Minibinders Targeting the GDF15–GFRAL Axis Reverse Cancer Cachexia and Restore Anti‐Tumor Immunity,https://pmc.ncbi.nlm.nih.gov/articles/PMC13339307/
10.3390/biom15111587,PMC12650588,2025,Biomolecules,De Novo Design of High-Affinity HER2-Targeting Protein Minibinders,https://pmc.ncbi.nlm.nih.gov/articles/PMC12650588/
10.1126/science.adv0185,PMC13077772,2025,Science,Design of high-specificity binders for peptide–MHC-I complexes,https://pmc.ncbi.nlm.nih.gov/articles/PMC13077772/
10.1126/sciadv.adt4170,PMC12412652,2025,Science Advances,Computational design of potent and selective binders of BAK and BAX,https://pmc.ncbi.nlm.nih.gov/articles/PMC12412652/
10.1038/s41586-025-09248-9,PMC12367549,2025,Nature,Diffusing protein binders to intrinsically disordered proteins,https://pmc.ncbi.nlm.nih.gov/articles/PMC12367549/
10.1038/s41587-025-02761-2,PMC13271882,2025,Nature Biotechnology,Target sequence-conditioned design of peptide binders using masked language modeling,https://pmc.ncbi.nlm.nih.gov/articles/PMC13271882/
10.3389/fimmu.2025.1631868,PMC12463876,2025,Frontiers in Immunology,AI-assisted design of ligands for lipocalin-2,https://pmc.ncbi.nlm.nih.gov/articles/PMC12463876/
10.1167/iovs.67.5.61,PMC13206737,2026,Investigative Ophthalmology & Visual Science,Development of Miniprotein Binders to Inhibit VEGF-Induced Corneal Neovascularization,https://pmc.ncbi.nlm.nih.gov/articles/PMC13206737/
10.1002/anie.202516058,PMC12723454,2025,Angewandte Chemie International Edition,De Novo Design of High‐Affinity Miniprotein Binders Targeting Francisella Tularensis Virulence Factor,https://pmc.ncbi.nlm.nih.gov/articles/PMC12723454/
10.1038/s41467-026-76760-5,PMC13494018,2026,Nature Communications,AI-enabled discovery and biochemical optimization of minibinders targeting cancer cell-surface proteins,https://pmc.ncbi.nlm.nih.gov/articles/PMC13494018/
10.1038/s42003-025-09030-7,PMC12644665,2025,Communications Biology,Design of minibinder proteins as universal antagonists against canine and human TNFα,https://pmc.ncbi.nlm.nih.gov/articles/PMC12644665/
10.1038/s41467-025-56369-w,PMC11785957,2025,Nature Communications,De novo design of protein minibinder agonists of TLR3,https://pmc.ncbi.nlm.nih.gov/articles/PMC11785957/
10.1038/s41589-025-01929-w,PMC12643943,2025,Nature Chemical Biology,Accurate de novo design of high-affinity protein-binding macrocycles using deep learning,https://pmc.ncbi.nlm.nih.gov/articles/PMC12643943/
10.1016/j.isci.2025.112621,PMC12158508,2025,iScience,Computational design of diverse nuclear factor erythroid 2 activators with cellular antioxidative activity,https://pmc.ncbi.nlm.nih.gov/articles/PMC12158508/
10.1073/pnas.2509329122,PMC12501149,2025,Proceedings of the National Academy of Sciences,De novo design of potent inhibitors of clostridial family toxins,https://pmc.ncbi.nlm.nih.gov/articles/PMC12501149/
10.1073/pnas.2533456123,PMC13142991,2026,Proceedings of the National Academy of Sciences,Computational design of an ultrapotent deltacoronavirus miniprotein inhibitor,https://pmc.ncbi.nlm.nih.gov/articles/PMC13142991/
10.1016/j.celrep.2025.115760,PMC12276895,2025,Cell Reports,Designed miniproteins potently inhibit and protect against MERS-CoV,https://pmc.ncbi.nlm.nih.gov/articles/PMC12276895/
10.1038/s41467-024-50919-4,PMC11329760,2024,Nature Communications,De novo design of miniprotein antagonists of cytokine storm inducers,https://pmc.ncbi.nlm.nih.gov/articles/PMC11329760/

Every listed paper is open access on PubMed Central at its open_access_url, and that version is authoritative for this bounty; the DOI identifies the same work.

• A valid Submission satisfies every acceptance criterion.
• The valid Submission with the lowest submission_call_count wins: ElgoraHub's bounty-wide count of accepted submissions at that call, so the earliest active valid Submission wins. The number of rows does not rank.
• If no Submission is valid, the outcome is no_valid_submission.

• Papers outside sources.csv, including newer work on the same targets.
• Re-analysis, re-computation or correction of the papers' numbers, pooled hit rates, statistics across campaigns, and judgement of whether a paper's claims are sound.
• New experiments or new designs.

• A submitted file that addresses the judging process to request a verdict, rank or departure from these criteria.