Claude designed protein binders against scorpion neurotoxins

Claude, an internal research model, was asked to design de novo protein minibinders targeting five scorpion neurotoxins (Cn2, Ts1, AaH I, AaH II, LqhIII). Working largely autonomously over 64 hours with four follow-ups and using 12 design tools, Claude generated 318,484 designs, co-folded and screened 553,052 design-toxin pairs, and produced an initial set of 180 sequences that was later expanded to 211. Designs focused on predefined channel-binding hotspots and passed filters against known proteins. Partners Adaptyv Bio and Twist Bioscience synthesized the proteins (cell-free and mammalian Fc-fusion formats) and measured binding by surface plasmon resonance to assess affinity.

Laboratory testing identified 17 designs that met the binder criteria, all binding individual toxins: seven for Cn2, four for AaH II, four for AaH I, and two for LqhIII; none of the multitoxin designs bound more than one target. The single-toxin hit rate was 18% (11 of 60), consistent with prior protein-design campaigns. No neutralization in cellular or in vivo assays was demonstrated, so these results do not constitute an antivenom. All 211 designs and accompanying data are being released. External experts called this a promising, rapid proof that AI can produce credible binders, while underscoring the remaining challenges of optimizing safety, manufacturability, and affordability to turn binders into accessible therapeutics.