Anthropic: Claude Designed 17 Protein Binders Against Scorpion Neurotoxins

What happened

The study began with a question from investment research firm Citrini Research: could a group of motivated nonexperts use AI to tackle a genuine problem in human health? Anthropic's life-science team brainstormed with Claude and landed on designing novel protein binders — small proteins that attach tightly to a target — against scorpion neurotoxins, hoping they could one day be used to develop a scorpion antivenom. Scorpions sting more than 1.2 million people every year, killing over 3,250, mostly children in regions with limited medical care — a neglected tropical disease.

Claude searched the Protein Data Bank and the AlphaFold database for toxin structures, selected binding hotspots, and used 12 design tools, while the team directed the work with four follow-up prompts. Claude selected 180 designs and the team added 31 more, for 211 total (60 single-toxin, 30 dual-toxin, 30 triple-toxin, 91 five-toxin). The two labs measured binding with surface plasmon resonance (SPR); the bar was evidence of binding in every replicate at both labs, a dissociation constant (KD) of 1 micromolar or tighter, and no signal against CLEC12A, an unrelated human protein.

The result: 17 designs met the criteria, all single-toxin binders — seven against Cn2, four against AaH II, four against AaH I, and two against LqhIII; none of the multi-toxin designs qualified. The strongest, Cn2-B1/B2, reached single-digit nanomolar affinity. Anthropic stated explicitly that it has not confirmed neutralization, has done no cell or animal experiments, and does not claim to have produced an antivenom; all 211 designs and binding data have been open-sourced.

Designing credible binders against five toxins in parallel, in a matter of days, is a great showcase of the AI-driven paradigm shift we are seeing in therapeutic discovery. This is especially true for neglected diseases such as animal envenomings, which have been largely starved of innovation due to low commercial incentives.

— Timothy P. Jenkins, Associate Professor, Technical University of Denmark (via Anthropic's blog)

Key facts

1. Design scale: 318,484 designs generated in 64 hours; 553,052 design-toxin pairs screened; on the order of 10,000 GPU hours used.
2. Toxin targets: Five scorpion neurotoxins — Mexico's Cn2, Brazil's Ts1, North Africa's AaH I and AaH II, the Middle East's LqhIII — all aimed at each toxin's channel-binding site.
3. Validation: 211 designs lab-tested, 17 met the criteria (all single-toxin binders); among the 60 single-toxin designs, 11 qualified (18%) — consistent with typical protein design campaigns.
4. Strongest binders: Cn2-B1/B2 bind Mexico's Cn2 toxin with single-digit nanomolar affinity; competition experiments suggest they attach on or near the same epitope as antibody LR, reported to neutralize Cn2 in mice.
5. Stated limits: Anthropic said it ran no cell or animal experiments and claims no antivenom; binding was tested against lab-grown recombinant toxins, not venom-purified toxins.
6. Open release: All 211 designs and their binding data have been released publicly for the research community.