A new study shows that using artificial intelligence to stabilize natural proteins before laboratory evolution produces dramatically better results than evolving from nature's original versions. Researchers at the Broad Institute discovered that AI-redesigned proteins serve as superior starting points for directed evolution, yielding proteins with better activity, specificity, and stability.

Why Does Starting Point Matter So Much in Protein Engineering?

When scientists want to create a protein with new abilities, they typically begin with an existing protein and evolve it in the laboratory over many generations, selecting for versions that gain desired functions. However, this process comes with a major tradeoff: laboratory-evolved proteins often become unstable because they sacrifice structural integrity to gain new capabilities. The Broad Institute team realized that if they started with a more stable protein, it would have more "stability to spare" to accommodate the mutations needed for new functions.

The researchers tested this hypothesis using botulinum neurotoxin proteases, enzymes best known as the active ingredient in Botox. They used an AI model called ProteinMPNN, developed by David Baker's laboratory, to redesign natural botulinum proteases with improved stability. The AI suggested new amino acid sequences that folded into the same 3D structure as the natural enzyme but with significantly greater structural robustness.

Then they applied PACE, a rapid laboratory evolution method pioneered by David Liu's group, to guide the AI-redesigned protein toward a new function: cutting ataxin-2, a protein involved in neurodegeneration. The results were striking. The new protein was 79 times better at cutting ataxin-2 than versions evolved from natural botulinum protease.

"The most important finding is that using AI to stabilize natural proteins can provide much better starting points for laboratory protein evolution than what we and other researchers have been using for decades. This insight could change the way researchers conduct protein evolution," said David Liu, senior author of the study and Richard Merkin Professor at the Broad Institute.

David Liu, Richard Merkin Professor and Director of the Merkin Institute of Transformative Technologies in Healthcare at the Broad Institute

How Does This Hybrid Approach Work Better Than Either Method Alone?

• AI Stabilization First: Artificial intelligence redesigns natural proteins to be more structurally stable without changing their core 3D shape, creating a more robust foundation for evolution.
• Laboratory Evolution Second: Scientists then use rapid evolution techniques like PACE to guide the stable AI-designed protein toward new functions, allowing it to accumulate beneficial mutations without collapsing.
• Synergistic Outcome: The combination produces proteins with superior activity, specificity, and stability compared to proteins evolved from natural starting points, because the extra stability provides flexibility for functional changes.

The team tested this approach across multiple botulinum proteases and different target substrates, and in every case, the AI-designed starting points outperformed their natural counterparts. Further experiments revealed something surprising: the activity-enhancing mutations that evolved in the AI-redesigned proteins could not be transplanted into the natural proteins without completely destabilizing them. This demonstrates that the AI-designed proteins were fundamentally different in their capacity to evolve.

"When proteins evolve new functions, they typically sacrifice stability in the process. That limits how much they can change during evolution. But if you start with a more stable protein, it has more stability to spare, so it can afford larger changes in pursuit of new functions," explained Nick Krasnow, a graduate student who led the research.

Nick Krasnow, Graduate Student at Broad Institute

What Could This Mean for Disease Treatment?

The implications extend far beyond botulinum proteases. The researchers note that this hybrid approach could be particularly valuable for engineering reverse transcriptases, enzymes whose stability often limits the effectiveness of prime editors, a cutting-edge gene-editing technology. Prime editors are being developed to treat a wide range of genetic diseases by precisely rewriting DNA sequences in living cells. If reverse transcriptase stability has been a bottleneck, this new strategy could unlock more powerful versions.

The study was published in Nature on July 22, 2026, and represents a fundamental shift in how protein engineers think about their work. Rather than viewing AI design and laboratory evolution as competing approaches, researchers can now see them as complementary tools that work best together. The findings suggest that neither AI design alone nor laboratory evolution alone achieves what their combination can accomplish.

The research team is now broadly applying this strategy to other protein types, with the goal of engineering improved versions of proteins used in medicine and biotechnology. The work was supported by the National Institutes of Health, the Howard Hughes Medical Institute, and the National Science Foundation.