Skip to content
← Back to Biology & Medicine

AI-designed proteins saved mice from a deadly snake toxin, given 15 minutes afterwards

Snakebite antidotes are still made from horse and sheep blood. Computer-designed proteins saved all five mice given one after a lethal dose.

5 of 5 mice survivedgiven the designed protein 15 minutes after a lethal toxin dose

Snakebite kills more than 100,000 people a year, and the treatment has barely changed in a century. Antivenom is still made by injecting horses or sheep with venom and harvesting antibodies from their blood. It is expensive, it needs refrigeration, and it works poorly against three-finger toxins, a family of small venom proteins that paralyse muscles or destroy tissue.

A team at the University of Washington and the Technical University of Denmark designed replacements on a computer instead, and published the results in Nature in January 2025. They used three AI tools in sequence: one to invent protein shapes that grip the toxin, one to pick the chemical sequence that would fold into each shape, and one to predict which designs would work before anything was made in a lab. About 2,000 designs were generated per target; only dozens were built and tested.

The designs blocked all three subfamilies of three-finger toxin in cell tests. In mice, the design aimed at short-chain nerve toxins protected all five animals even when injected 15 minutes after a lethal dose, and again at 30 minutes. The design for long-chain toxins saved 3 of 5 mice at 30 minutes. Mice given the proteins on their own were unharmed over two weeks.

What it means

Because these are small, stable proteins, they can be grown in microbes rather than drawn from animal blood, which the authors argue could make treatment cheaper and easier to store in the rural places where most snakebites happen. That is the promise, not the present: this was purified toxin in mice, so no treatment has changed for anyone bitten today. It is also a sign of what designed proteins may do for diseases that get little commercial attention.

Caveats

  • Mice, in groups of five. Those group sizes are too small to measure anything but a large effect.
  • The tests used purified toxins, not whole venom. Real venom is a mixture, and the authors note that neutralising it would need several designed proteins together.
  • Only the three-finger toxin family is covered. Vipers, which cause much of the global snakebite burden, use different toxins.
  • No human testing, and no sign yet of the safety, dosing and manufacturing work a product would need.
  • Protection was partial in some conditions, for example 3 of 5 mice for the long-chain binder given 30 minutes after the toxin.