Designing de novo Snake Antivenoms using Generative AI

Aim: Aim: Develop shelf-stable, region-specific recombinant antivenoms for Pakistani snake species using AI-driven protein design.

Potential impact: Address the 40,000-50,000 annual snakebite cases and 8,200 deaths in Pakistan. Replace expensive imported antivenoms with locally-manufacturable alternatives, reducing costs while eliminating refrigeration dependency critical for rural areas.

Previous work: As a proof of concept, we employed existing approaches to design de novo binders against Russell's viper PLA₂ toxin, achieving reasonable structural confidence (pLDDT 94, pTM 0.87) with sub-micromolar binding affinity (Kd ~1.6 µM). 

Method: Create consensus venom peptides from Pakistan's four most common venomous snakes, then deploy RFdiffusion for backbone generation, ProteinMPNN for sequence optimization, and AlphaFold2/Rosetta validation to generate 40 top binder candidates per target. Express and purify selected candidates using Gibson assembly and E. coli expression systems.

Current progress: Currently deploying advanced diffusion-based protein design tools and optimizing computational pipeline for large-scale, multi-target binder generation.

Next Steps: Generate consensus peptides, screen top 3 highest-affinity binders, and validate through recombinant expression and in vitro neutralization assays over 18 months.