Anthropic reports that an internal research model designed protein "minibinders" against scorpion neurotoxins, and that lab tests at two contract research organizations confirmed 17 of 211 designs as binders. The designs and their binding data are released as a Hugging Face dataset. Anthropic stresses that binding was measured only against recombinant toxins, so it does not claim to have produced an antivenom.
Why scorpions
Scorpions sting 1.2 million people or more each year, killing more than 3,250, mostly children in regions with limited medical access. Existing antivenoms come from the antibodies of immunized horses, a process that takes months. They usually cover only a few neurotoxins and are costly or scarce in many places. Treatments are unlikely to earn back their development cost, which is why envenomation is described as a neglected tropical disease. The dangerous toxins attach to sodium channels in nerve cells and make nerves misfire, which can lead to heart failure and fluid in the lungs.
The project followed August's single-prompt binder study. A team at Citrini Research, a financial investment research firm, had asked whether Claude could let motivated nonexperts tackle a real health problem, and Anthropic's life-science team picked scorpion antivenom with Claude's help.
The campaign
Claude started from the earlier study's prompt, modified with Citrini, and ran for 64 hours inside Claude Science, mostly autonomously, with four follow-up prompts. It chose structures from the Protein Data Bank and the AlphaFold database and five hotspots per toxin, targeting five toxins from four regions: Cn2 (Mexico), Ts1 (Brazil), AaH I and AaH II (North Africa) and LqhIII (Middle East). The toxins share one fold but are only 26% to 62% identical in sequence, making them a test of designs that bind several toxins at once, which would cut the cost of a universal cocktail.
- Scale: 12 design tools generated 318,484 designs, filtering out anything resembling a toxin or known protein. Claude co-folded and screened 553,052 design-toxin pairs, on the order of 10,000 GPU hours, and picked 180 designs. Most, including the strongest binder, Cn2-B1, came only after the third follow-up.
- The 211: Claude helped the team add 31 more, giving 60 single-toxin designs (12 per toxin), 30 against the two β-family toxins, 30 against the three α-family toxins and 91 against all five. Claude emailed the sequences to Twist Bioscience.
- Testing: Adaptyv Bio expressed designs in a cell-free system and Twist Bioscience in human kidney cells as antibody-Fc fusions, and both measured binding by surface plasmon resonance. A binder had to hit at both labs in every replicate with KD of 1 µM or tighter, and show no signal against the unrelated human protein CLEC12A.
Results
17 of 211 designs met the criteria, each against one toxin: seven against Cn2, four against AaH II, four against AaH I and two against LqhIII, and none against Ts1. Eleven of the 60 single-toxin designs were binders (18%), in line with typical protein design campaigns, so the other six binders came from multitoxin designs but each bound only one toxin. No multitoxin design met the criteria for more than one toxin. Anthropic cites its prompting, tool limits and the biology itself as possible reasons.
The two strongest, Cn2-B1 and Cn2-B2, bound Cn2 in the single-digit nanomolar range at both labs. Three reference antibodies and nanobodies are reported below 1 nM against Cn2 or AaH II and protect mice, so antibodies remain the bar to beat. In competition experiments, both Cn2 binders attached on or near the epitope of antibody LR, reported to neutralize Cn2 in mice, which suggests they would likely neutralize it too. Nothing was tested in cells or animals.
What comes next
Advancing Cn2-B1 would need confirmation against toxin purified from venom, patch-clamp recordings of sodium-channel current, and mouse protection tests, then safety, stability, affordability and access. Outside experts Susana Vázquez Torres (CNIO, Madrid) and Timothy P. Jenkins (Technical University of Denmark) called it a promising first step. Vázquez Torres says binding is no longer the main bottleneck, and the real impact will come from the parts of the pipeline beyond binding. It follows Anthropic's life-sciences lab opening.