Claude designs scorpion-toxin binders on its own: 17 of 211 designs bind in two independent labs

Scorpion neurotoxins now have AI de novo-designed protein binders whose binding was confirmed at two contract research labs, covering five toxins from scorpions in Mexico, Brazil, North Africa and the Middle East. De novo-designed proteins had previously neutralized snake neurotoxins and protected mice, but no de novo-designed binders against scorpion neurotoxins had been described; existing scorpion antivenoms rely on immunized animals, take months to make and usually cover only a few neurotoxins. All 211 designs were measured for binding by surface plasmon resonance at Adaptyv Bio, in a cell-free system, and at Twist Bioscience, expressed in human kidney cells as Fc fusions; a design counted only if it was a hit at both labs, bound in every replicate with a dissociation constant of 1 µM or tighter, and showed no signal against the unrelated human protein CLEC12A. Seventeen passed.

Claude ran the campaign with little human direction: it chose toxin structures from the Protein Data Bank and the AlphaFold database, picked five hotspot residues per toxin, and set its own quality thresholds and filtering rules. The earlier single-prompt experiment showed one-shot design; this campaign ran for 64 hours, with the team steering it through four follow-up prompts, and most designs, including the strongest binder Cn2-B1, were selected only after the third follow-up. Claude used 12 design tools to generate 318,484 designs, filtered out sequences resembling toxins or known proteins, co-folded and screened 553,052 design-toxin pairs, and returned 180 designs, using on the order of 10,000 GPU hours.

The two strongest binders, Cn2-B1 and Cn2-B2, target Cn2 from the Mexican scorpion Centruroides noxius and reached single-digit nanomolar affinity at both labs. Antibodies and nanobodies remain the standard for de novo designs to beat, with three molecules reported below 1 nM against Cn2 or AaH II and shown to protect mice from scorpion toxins. Competition experiments indicate both binders attach on or near the same epitope as antibody LR, whose binding site covers the part of Cn2 that attaches to the sodium channel; neither binder has been tested in cells or live animals.

The results cover binding to lab-grown recombinant toxins only; no design was tested for neutralization in cells or for protection in live animals, so the work does not claim an antivenom. The stated path for advancing one design is to confirm it also binds toxin purified from venom, then assess function in vitro with patch-clamp recordings of sodium-channel currents, then test protection in mice. All 211 designs, predicted structures, scores and experimental data are released for others to build on.

Cn2-B1 and Cn2-B2 were tested only in competition experiments; sharing an epitope with LR suggests they will likely neutralize Cn2, but that has not been confirmed in cells or animals. No multitoxin design met the criteria for more than one toxin, and the source lists prompting, limits of available design tools and the biological feasibility of multitoxin design as possible explanations without settling on one. Binding was measured against recombinant toxins, not toxin purified from venom.