Anthropic's Claude Designs Protein Binders to Target Scorpion Neurotoxins

At a glance

Anthropic asked its AI, Claude, to design small proteins that bind to scorpion toxins. Over about 64 hours, the AI did most of the work on its own, while researchers gave it four additional requests to steer the direction. After lab testing, 17 of the 211 designs bound well to the toxin.

Key points

1. 1

   Claude generated more than 310,000 protein design candidates and selected 180 of them.

2. 2

   Two outside research labs tested the 211 designs in the lab.

3. 3

   17 designs passed the criteria for binding to the toxin, but the team has not yet confirmed whether they actually neutralize the poison.

4. 4

   The researchers released all 211 designs and their data so other researchers can use them.

Why it matters

Many people are stung by scorpions every year, but treatment development has received little investment, leaving the problem overlooked. This case shows whether AI can quickly identify starting points for developing such treatments.