Anthropic’s Claude AI designs protein binders for scorpion neurotoxins

Anthropic has released results from an AI-driven research campaign in which its Claude model designed potential protein binders to target scorpion neurotoxins. During a 64-hour period, the model generated 318,484 designs targeting five specific neurotoxins associated with scorpion species from Mexico, Brazil, North Africa, and the Middle East.

Out of 211 candidates selected for laboratory testing—180 chosen by the AI and 31 added by researchers—17 met the predefined criteria for binding to the target toxins. The successful designs targeted four of the five neurotoxins, including the Cn2 toxin from Mexico. The two strongest binders for Cn2 showed high binding affinities in the single-digit nanomolar range.

While the results are a significant milestone for AI-assisted drug discovery, researchers emphasized that these proteins are not yet antivenoms. The study confirmed that the proteins can physically attach to the toxins, but it did not test whether they can neutralize the venom's effects in cells or living organisms, nor was their safety evaluated. The laboratory testing was conducted by Adaptyv Bio and Twist Bioscience.