#00139
Case study of
#00139 Computationally design de novo toxin-neutralizing proteins with deep learning
Implementer
Institute for Protein Design / Baker Lab (University of Washington, HHMI); Technical University of Denmark (Laustsen, Jenkins); Liverpool School of Tropical Medicine (Casewell); University of Northern Colorado (Mackessy)
Location
Description
Using the deep-learning method RFdiffusion (with ProteinMPNN sequence design and AlphaFold2 filtering), researchers de novo designed small (~100-amino-acid) proteins to bind three-finger toxins from elapid venom: short-chain α-neurotoxins (design "SHRT"), long-chain α-neurotoxins/α-cobratoxin ("LNG"), and cytotoxins ("CYTX"). Crystal structures closely matched the computational models. The neurotoxin binders were sub-nanomolar and highly thermostable and fully neutralized their targets in patch-clamp assays; in mice they gave complete protection against lethal neurotoxin challenge, including as post-envenoming rescue. The cytotoxin binder neutralized cytotoxicity in vitro but did not reduce dermonecrosis in vivo. The authors position the binders as low-cost, animal-free antivenom components or "fortifying agents." Published in Nature (Jan 2025).
Metrics
4Lessons learned
Documented Jul 8, 2026