AI-redesigned starting points and outcomes enhance protein evolution - Nature

What the report says

Nature reports that researchers established a workflow combining AI-based protein sequence redesign with laboratory evolution to improve enzymes beyond what was achieved when evolution began from natural proteins. The study used ProteinMPNN, a protein-design model, to redesign three botulinum neurotoxin proteases, producing variants that retained full catalytic efficiency while showing improved stability.

The researchers then compared redesigned enzymes with their wild-type counterparts in phage-assisted continuous evolution campaigns. According to the article, across four campaigns, the AI-redesigned starting points consistently led to proteases with higher activity than those evolved from wild-type proteins under the same selection conditions. The authors report that redesign made the enzymes more tolerant of later mutations, allowing access to high-performing sequences that did not work when placed back into wild-type enzyme backgrounds.

In one test involving botulinum neurotoxin E protease, the team evolved both the natural and AI-redesigned versions to cut ataxin-2, a protein described as therapeutically relevant. Proteases derived from the redesigned enzyme reached higher catalytic efficiency and stability while reducing cleavage of the native substrate. Nature reports that the best redesigned-lineage variant showed more than 79-fold greater selected specificity for ataxin-2 than the best variant evolved from the wild-type starting point.

The finding matters because directed evolution often begins with natural proteins that can lose stability as mutations accumulate. The reported approach suggests computational redesign can give enzymes a more robust starting position before experimental evolution, potentially helping protein engineers develop more stable, selective or active enzymes for biomedical research, therapeutics, synthesis and environmental applications.