Baker Lab Uses AI to Design Enzymes That Cut Specific Proteins

Researchers led by Nobel Prize winner David Baker’s lab at the University of Washington have used AI tools to design new enzymes that cut chosen proteins at specific points. The work is described in two papers published in Nature on 14 October 2026. These are lab results, not medical treatments.

The enzymes are proteases, which break the peptide bonds that hold proteins together. Those bonds are very stable on their own; as the cysteine paper puts it, they “are stable for hundreds of years in neutral aqueous solution.” Designing a protease from scratch has long been a hard problem.

Two kinds of designed protease

The first paper, led by Hojae Choi, describes proteases built around an activated cysteine, a type of amino acid. The team reports “a success rate of 13/69 = 19%”, meaning 13 of 69 designs worked, with reactions sped up by “up to 3 × 10^7” times compared with no enzyme. The authors say the designs have folds “very different from proteases in nature”, and that “the designed proteases efficiently cleave their targets in mammalian cells.”

The second paper, led by Anqi Chen and Kejia Wu, describes proteases built around zinc. It reports that “of 135 designs tested, 36% catalyse cleavage of their cognate peptide substrates at the targeted scissile bond.” The best design sped up the reaction by more than 10^8 times, and four changes pushed this to more than 10^10 times, according to the paper.

The team also designed zinc proteases that “selectively cleave the human disease-associated proteins TDP-43, amyloid-β and serum amyloid A”. These proteins are linked to conditions including dementia and Alzheimer’s disease. Ars Technica reported that the designed enzymes were generally less active than the natural enzymes used as starting points.

Not a treatment yet

Ars Technica noted that using such enzymes as medicines is “complicated”. Proteins are hard to deliver to where they are needed in the body and can trigger an immune response. Targets inside cells would be even harder to reach.

The metalloprotease paper notes that provisional patent applications have been filed on aspects of the work.

What this means

• AI can now design enzymes that cut. This was a long-standing challenge in protein design.
• The targets can be chosen. The enzymes can be aimed at specific proteins, including disease-linked ones.
• It is early research. These are not treatments, and delivery remains a big hurdle.

Sources

1. Choi, H. et al., “De novo design of cysteine proteases”, Nature, 14 October 2026. https://www.nature.com/articles/s41586-026-11143-w
2. Chen, A., Wu, K. et al., “Computational design of metalloproteases”, Nature, 14 October 2026. https://www.nature.com/articles/s41586-026-11167-2
3. Ars Technica, “Researchers use AI to design proteins that cut up other proteins”, John Timmer, 14 October 2026. https://arstechnica.com/science/2026/10/researchers-use-ai-to-design-proteins-that-cut-up-other-proteins/

Related stories

• Generate Biomedicines and Zealand Pharma Team Up to Design Obesity Protein Drugs With AI
• The Invisible Signature: Google DeepMind Has Started Watermarking AI-Designed Proteins
• The Drug Was Designed by an AI: How Insilico and Lilly Just Changed Medicine Forever