Researchers at the Broad Institute of MIT and Harvard have demonstrated that integrating artificial intelligence protein design with traditional laboratory evolution significantly accelerates the engineering of stable, high-function enzymes. The study, published in Nature, reveals that beginning evolutionary processes with AI-optimized protein scaffolds yields superior outcomes compared to using naturally occurring counterparts, a finding poised to reshape methodologies in synthetic biology and therapeutic development. Led by David Liu, senior author and director of the Broad Institute’s Merkin Institute, alongside graduate student Nick Krasnow, the team focused on repurposing botulinum neurotoxin proteases. Rather than starting with native bacterial enzymes, the researchers employed ProteinMPNN, an AI model developed at the University of Washington, to redesign the proteases for enhanced structural stability while preserving their native three-dimensional folding. These AI-optimized variants were subsequently subjected to Phage-Assisted Continuous Evolution, or PACE, a rapid lab evolution technique pioneered by Liu’s laboratory, to acquire the ability to cleave ataxin-2, a misfolded protein implicated in neurodegenerative diseases. The hybrid approach produced enzymes with markedly improved performance. The AI-evolved proteases demonstrated a seventy-nine-fold increase in cleavage efficiency against ataxin-2 compared to proteins evolved from natural starting materials. Furthermore, the AI-designed scaffolds consistently outperformed natural proteins across multiple enzyme-target combinations. Mechanistic analysis indicated that the additional structural stability inherent in the AI-designed proteins provided an evolutionary buffer. While functional mutations typically compromise protein stability, the AI-optimized variants retained excess structural robustness, allowing them to absorb the destabilizing effects of functional adaptation without losing folding integrity. In contrast, attempting to transplant these same functional mutations into natural proteins resulted in complete structural collapse. The study underscores that neither computational design nor laboratory evolution alone achieves optimal results when tackling complex enzymatic repurposing. By decoupling stability engineering from functional innovation, researchers can bypass the historical trade-off that has long constrained directed evolution. Liu emphasized that establishing superior starting materials is a critical determinant of evolutionary success, noting that this synergistic framework could rapidly transform protein engineering pipelines. The Broad team has already extended the strategy to other enzyme classes, including reverse transcriptases used in prime editing systems, where stability constraints previously limited therapeutic optimization. As AI-driven structural biology matures, this validated protocol offers a scalable blueprint for designing next-generation therapeutics, industrial biocatalysts, and genetic editing tools. By ensuring engineered proteins remain robust while acquiring novel functions, the combined methodology addresses a fundamental bottleneck in translational biotechnology and establishes a new standard for precision protein engineering.