Claude Science designs protein candidates against scorpion toxin

What the lab results show

Protein candidates are small molecules designed to bind to a specific target. Claude selected 180 candidates; the research team added 31. Two of the most promising bound to Cn2 at the same site as an antibody known to neutralize the toxin in mice, or at a nearby site.

The work was conducted in Claude Science, Anthropic’s environment for researchers. Independent laboratories confirmed the results. External experts Susana Vázquez Torres and Timothy P. Jenkins called the work a promising first step.

Binding is not treatment

Anthropic’s earlier study offers context: Claude designed mini-proteins that bound to 14 of 15 target proteins. The share of successful candidates was 22% to 35%, compared with a typical rate of 10% to 15%.

I think the important distinction here is between designing a protein that binds and demonstrating that it works as an antidote. The results establish the first; the second has not been tested. The announcement leaves unanswered whether these candidates can neutralize Cn2 in an animal, and whether binding at the observed sites is enough to make that possible.

That gap sets the stakes for what comes next: the lab can identify promising designs, but the therapeutic claim still depends on evidence the study does not yet provide.