Claude designs proteins to investigate scorpion toxins
Anthropic used a Claude research model to design proteins intended to bind to scorpion toxins. Initial laboratory tests identified 17 candidates, but have not shown that they neutralise venom.
Summary
A Claude research model helped design small proteins intended to bind to scorpion neurotoxins, in an Anthropic experiment exploring AI’s role in drug discovery.
Of 211 designs tested in laboratories, 17 met the team’s criteria for binding to a toxin. This is an early research result: the team has not confirmed that the proteins neutralise venom and does not claim to have produced an antivenom.
In practice
The team asked Claude to work on five scorpion neurotoxins from different regions. The model selected structures and generated candidates, with additional direction from researchers during the process.
Adaptyv Bio and Twist Bioscience tested the designs. The 17 successful candidates each bound to one of four toxins; none of the tested designs met the criteria for binding to more than one toxin.
Context
Existing antivenoms can be effective when administered early, but their production and distribution remain difficult in some regions. Research into computationally designed proteins explores other possibilities for future treatments.
Anthropic has released all 211 designs and their data. The most promising candidates still need further testing to establish whether they neutralise toxins and could eventually contribute to safe, accessible treatments.
Why it matters
- The experiment combines AI-assisted design with laboratory testing.
- The released data allows other teams to examine the results.
- Binding to a toxin is not the same as neutralisation or clinical effectiveness.