Anthropic says an internal research version of its Claude AI has designed proteins that grab onto the nerve toxins in scorpion venom, a possible first step toward new treatments for stings. The company says Claude worked mostly on its own for 64 hours, with four follow-up instructions from Anthropic's team. Anthropic wrote those instructions with the investment research firm Citrini Research, which had asked whether AI could help a group of motivated non-experts take on a real health problem.
The designs are binders: small proteins that attach tightly to one specific target, the way antibodies do in the body. The scorpion toxins most dangerous to people latch onto sodium channels, proteins in the walls of nerve cells that let them fire. A binder that covers the part of a toxin that latches on could, in principle, stop the toxin from reaching the channel.
Scorpions sting 1.2 million people a year or more and kill more than 3,250, mostly children in places with little medical care, Anthropic says. Most of today's antivenoms are made from antibodies in the blood of animals injected with venom, usually horses, which can take months and usually works against only a small number of toxins.
Claude aimed at five toxins from four kinds of scorpion found in Mexico, Brazil, North Africa and the Middle East. It produced 318,484 designs on computers and picked 180, and the team added 31 more with its help. Two contract labs Anthropic hired, Adaptyv Bio and Twist Bioscience, made the 211 designs and tested whether they held onto their toxins, though the designs for one toxin were tested at only one of the labs. Seventeen passed Anthropic's test for a binder, each against a single toxin. Of the designs made to grab one toxin, 11 of 60 passed, which Anthropic says is a typical rate for protein design.
None of the designs meant to grab several toxins at once held onto more than one. Those could have cut the number of molecules a single, universal antivenom would need. Anthropic says the reasons could include its own instructions, the design tools, or whether such a protein is possible at all.
The two strongest designs hold tightly to Cn2, a toxin from a Mexican scorpion. Anthropic's tests suggest they attach at or near the same spot as an antibody reported to neutralize that toxin in mice, which the company says makes it likely they would neutralize it too.
Anthropic says it has not made an antivenom. The designs were tested only against toxins grown in the lab, not taken from venom, and none has been tried on cells or in animals. Two outside scientists Anthropic asked to review the work, Susana Vázquez Torres of Spain's National Cancer Research Centre and Timothy P. Jenkins of the Technical University of Denmark, called it a promising first step, and Vázquez Torres said the harder part is turning such proteins into drugs that are safe, affordable and widely available. It is not yet known whether any of the 17 will block a toxin in a living animal, or whether funders would carry one as far as patients.