Anthropic’s Claude Yields 17 Scorpion Toxin Binders in Lab Tests

Anthropic’s Claude yielded 17 protein designs that met laboratory criteria for binding scorpion neurotoxins in results released Oct. 20, an early drug discovery finding that does not establish whether the designs can neutralize venom.

An internal model generated 318,484 designs during a 64-hour campaign and selected 180 for testing. Anthropic’s life sciences team added 31 with Claude’s help, bringing the total to 211. Working within Claude Science, the model operated mainly on its own under team supervision, guided by an initial prompt and four follow-ups.

The experiment targeted five neurotoxins: Cn2 from Mexico, Ts1 from Brazil, AaH I and AaH II from North Africa, and LqhIII from the Middle East. Seven designs met the binding criteria for Cn2, four for AaH I, four for AaH II and two for LqhIII. None qualified for Ts1.

Adaptyv Bio and Twist Bioscience carried out the binding tests. Each of the 17 qualifying designs bound a single toxin. Claude also produced designs intended to attach to multiple toxins, but none met the criteria for binding more than one. The tests measured whether the proteins attached to their targets, not whether they stopped the toxins’ effects.

The two strongest Cn2 binders, Cn2-B1 and Cn2-B2, showed binding affinities in the single-digit nanomolar range, a measure of how tightly they attached. In competition tests, they appeared to bind at or near the site targeted by an antibody shown to neutralize Cn2 in mice. That result does not demonstrate that either Claude-designed binder neutralizes Cn2.

The tests used toxins produced in a lab rather than toxins purified from venom. The strongest binders have not been tested in cells or animals, so their ability to block a toxin’s effects and their safety remain unestablished. Binding to a laboratory target is one step in evaluating a possible treatment, but these results do not constitute an antivenom.

The campaign adds a laboratory example to an AI drug discovery investment thesis examining whether faster candidate design could increase work for testing laboratories and research equipment providers. Here, outside labs tested a small selection from hundreds of thousands of computer-generated designs. Anthropic is releasing all 211 designs and their binding data for other researchers.