How Claude designed protein binders against scorpion neurotoxins
Short summary
Anthropic had an internal Claude model design small proteins meant to latch onto scorpion neurotoxins, then had outside labs test them. Seventeen of 211 designs bound their targets, a possible early step toward cheaper antivenom, though none have been shown to neutralize venom yet.
- Claude generated about 318,000 candidate designs and picked 180 for testing, mostly running autonomously over 64 hours with four human follow-up prompts.
- Of 211 tested designs, 17 bound one of four dangerous scorpion neurotoxins. None of the designs aimed at several toxins at once worked for more than one.
- Anthropic says binding is no longer the bottleneck; the hard part is turning binders into safe, affordable drugs, and they haven't tested whether these neutralize venom in cells.
Full summary
Anthropic used an internal research model, running inside Claude Science, to design small proteins called minibinders against four scorpion neurotoxins that are dangerous to humans. The goal was to see whether AI could help with a neglected problem: scorpion stings hit roughly 1.2 million people a year and kill more than 3,250, mostly children in regions with thin medical access. The work started from a prompt developed with Citrini Research, then ran for 64 hours with only four human follow-ups. Claude analyzed data, made figures, and drafted the report. Out of 318,484 generated designs, it picked 180 for testing, and the team added 31 more. Adaptyv Bio and Twist Bioscience lab-tested all 211, and 17 bound their targets. The team is releasing all 211 designs and their data.
The targets matter. The toxins that cause the worst harm bind to sodium channels in nerve cells, and disrupting those channels triggers a surge of stress chemicals that can lead to heart failure and fluid in the lungs. Current antivenoms are made by milking thousands of scorpions, immunizing horses, and purifying antibodies from their blood, a process that takes months and works well against only a few neurotoxins. The five toxins chosen came from Mexico, Brazil, North Africa, and the Middle East, and were picked to test multitoxin design. They share a similar overall shape but are only 26 to 62 percent identical in sequence. Claude selected the protein structures itself from the Protein Data Bank and AlphaFold database, and chose five hotspot positions per toxin to target. A single binder that hits several toxins would cut the number of molecules needed for a broad antivenom cocktail.
The limits are stated plainly. The team has not shown that any of the 17 binders neutralize venom in cells, so this is not an antivenom. Outside experts Susana Vázquez Torres and Timothy P. Jenkins called it a promising first step, but both pointed out that the hard part now is turning binders into safe, cheap, widely available drugs. Binding is no longer the bottleneck. Manufacturing, stability, and affordability are.