Claude-designed protein binders against five scorpion neurotoxins
A generative design campaign run by an internal Anthropic model produced 318,484 candidate proteins against five scorpion-venom neurotoxins; of the 211 that were synthesised and tested, 17 met the binding criterion at both contract laboratories, the tightest at single-digit nanomolar affinity, and none was tested for venom neutralisation.
- Lab
- Anthropic
- Model
- Internal Anthropic research model operating within Claude Science, not named or versioned in the post
- Field
- Biology
- Date
- 2026-10-20
What was found
Anthropic's initial prompt named five neurotoxins from four regions, Cn2 from Centruroides noxius in Mexico, Ts1 from Brazil, AaH I and AaH II from North Africa and LqhIII from the Middle East, along with the protein design and folding tools to use and the rules for novelty, scoring and filtering. The model chose the target structures from the PDB and the AlphaFold database, picked five hotspots per toxin, set its own quality thresholds, drove 12 design tools to 318,484 designs and co-folded 553,052 design-toxin pairs, using on the order of 10,000 GPU hours. It selected 180 designs for synthesis 64 hours after the first prompt, and 31 more were added with its help, for 211 tested in total. Of those, 17 met the hit criterion: binding in every replicate at both laboratories with every dissociation constant at 1 micromolar or tighter, and no signal against an off-target stickiness control. All 17 bound a single toxin, seven against Cn2, four against AaH II, four against AaH I and two against LqhIII, with Cn2-B1 and Cn2-B2 in the single-digit nanomolar range, while every design intended to bind more than one toxin failed.
Novelty check
Searched Europe PMC on 2026-10-22 for de novo binders against scorpion toxins ("de novo" AND scorpion AND (binder OR "protein design"), 87 hits) and for the prior art the post itself cites ("de novo designed proteins" AND neurotoxin, 23 hits). The closest published work is Vazquez Torres et al., "De novo designed proteins neutralize lethal snake venom toxins", Nature, January 2025 (doi 10.1038/s41586-024-08393-x), which reached neutralisation and protected mice against snake three-finger toxins; no published de novo binder campaign against scorpion alpha or beta neurotoxins turned up. Anthropic's post cites that paper as the earlier effort, and its first author, Susana Vazquez Torres, is one of the two outside experts Anthropic quotes reviewing this work, so the nearest prior art and part of the review are not independent of each other. No prior registry entry covers scorpion toxins; 2026-09-23-array-associated-reverse-transcriptases is the registry's other Anthropic agent-run biology result and concerns phage genomics.
Caveats and known objections
Entered at claimed under the rule that lab announcements start there. The evidence is binding and nothing else: surface plasmon resonance against recombinant toxins, run at two contract research organisations commissioned by Anthropic, Adaptyv Bio by cell-free expression and Twist Bioscience in mammalian cells as Fc fusions, so the two-laboratory requirement is a paid second vendor rather than an independent check, and LqhIII was assessed at Adaptyv Bio only. The post states that no design was tested for neutralisation in vitro or for efficacy in vivo and that it does not claim to have produced an antivenom. A competition experiment places the Cn2 binders on or near the epitope of antibody LR, which has been reported to neutralise Cn2 in mice; that is suggestive and is not a neutralisation result. Binding was shown against recombinant toxins rather than venom-purified ones. All multi-toxin designs failed, and the post offers its own prompting, the limits of the available tools and the biological feasibility of multi-toxin design as candidate explanations. Autonomy is graded search-scaffold on the strictest defensible reading: the humans' prompt fixed the targets, the tool set and the rules for novelty, scoring and filtering, which is the human-designed loop that grade describes, even though the model chose hotspots, thresholds and which designs to synthesise. A reading that weighted those choices would reach ai-assisted, and the post's own framing, that the model worked from a prompt and four follow-ups under supervision and then analysed the data and drafted the report, is why the weaker grade was taken. No model name or version is given and no Anthropic authors are named on the post, so the only named people are the two outside reviewers, who are not collaborators and are therefore not listed here. The technical report PDF exceeded the fetch size limit and was not read for this entry, so every count and criterion above comes from the post and the released dataset.
Nobody outside the lab has checked this yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Claimed and Search scaffold.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded claimed for verification and search scaffold for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Claude-designed protein binders against five scorpion neurotoxins. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-20-claude-scorpion-toxin-bindersBibTeX
@misc{whataifound-anthropic-2026-binders,
title = {Claude-designed protein binders against five scorpion neurotoxins},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Anthropic. Verification: Claimed. Autonomy: Search scaffold.},
url = {https://whataifound.org/finding/2026-10-20-claude-scorpion-toxin-binders}
}