Claude designed protein binders against scorpion neurotoxins — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Claude generated 318,484 protein designs over ~64 hours of largely autonomous work, selecting 180 for testing while human team members added 31 more, producing 211 designs that partners Adaptyv Bio and Twist Bioscience then lab-tested.
- Of the 211 tested designs, 17 met binding criteria — each against a single scorpion neurotoxin — while none of the designs meant to bind multiple toxins simultaneously passed criteria for more than one target.
- The campaign targeted five neurotoxins from scorpion species across Mexico (Cn2), Brazil (Ts1), North Africa (AaH I, AaH II), and the Middle East (LqhIII), with Claude choosing toxin structures from the Protein Data Bank and AlphaFold databases.
- The computational footprint came to roughly 10,000 GPU hours across 12 design tools, with Claude running the work after an initial prompt and four follow-ups, according to the technical report.
- Scorpion stings affect at least 1.2 million people annually and kill more than 3,250 — mostly children in regions with limited medical access — which is why the team selected this as a neglected-disease test case.
- Outside reviewers Susana Vázquez Torres (CNIO) and Timothy P. Jenkins (Technical University of Denmark) called the work a promising first step, with Jenkins highlighting the speed-up for envenomation research and Torres emphasizing the bigger challenge of turning binders into safe, affordable drugs.
- Anthropic and Citrini Research co-developed the initial prompt and follow-ups, and Anthropic is releasing all 211 designs plus experimental data as a public resource for the field.
Why it matters: By releasing all 211 designs and their lab-test data, the team gives the field a public dataset on AI-generated venom binders — 17 of 211 hit their target — but Torres and Jenkins both flag in the piece that binding was never the hard part, and turning these into deployable antivenoms for the 3,250+ killed annually remains the unsolved downstream work.
Ask SkimNews
