Research watchScienceNew
AI-designed enzymes cut chosen protein targets, including ones linked to Alzheimer's and ALS, in lab tests
Two Nature papers from the Institute for Protein Design report proteases designed from scratch with AI that cut chosen sites in target proteins. Most designs were far weaker than natural enzymes, and none has been tested as a treatment.
Evidence: 5 sources · 1 official record · 3 independent reports · 1 research paper · 3 open questions
What happened
- On 14 October, Nature published two papers from David Baker's lab at the University of Washington describing proteases designed de novo with AI tools.
- One paper designs zinc-based metalloproteases; the other designs cysteine proteases, which use an activated cysteine to cut the protein backbone.
- The designs targeted sites in proteins including amyloid-beta, serum amyloid A and TDP-43, which are linked to Alzheimer's disease, amyloidosis and ALS.
- In the cysteine protease paper, 13 of 69 tested designs (19%) were active; in the metalloprotease paper, reports say more than a third of tested designs cut their target.
Why it matters
Proteases that can be aimed at a chosen site would be a new kind of research tool and, eventually, a possible drug format, because one enzyme can cut many target molecules. Designing enzymes that catalyze reactions has been much harder than designing proteins that simply bind.
What is confirmed and what is claimed
Confirmed
- Both papers are peer-reviewed and were published in Nature on 14 October (Nature; UW Medicine; also reported by C&EN and GeekWire). 1234
- The team used RFdiffusion-based generative design methods to build the enzymes' active sites and binding pockets (papers and coverage).
- Tests were on purified proteins and peptides, and in one study on a target inside cultured human cells; there were no animal or human trials (UW Medicine; Phys.org). 25
Claimed, not independently verified
- The researchers suggest catalytic designer proteases could one day treat disease at lower doses than drugs that bind and block a target. This has not been tested.
- Reported rate enhancements, such as up to roughly 30 million-fold for the cysteine proteases relative to the uncatalyzed reaction, come from the papers and have not been independently reproduced.
Proteases are enzymes that cut other proteins. Nature uses them everywhere, but each natural protease cuts at the sites it evolved to recognize. The two new papers, from the Institute for Protein Design at the University of Washington, describe proteases built from scratch to cut sites the researchers chose.
The team used generative design software from the RFdiffusion family to lay out an active site and a pocket that holds the target sequence, then used other tools to pick amino-acid sequences likely to fold into that shape. One paper builds enzymes around a zinc ion. The other uses a cysteine as the cutting residue. Separately, the metalloprotease paper describes designed proteins, such as caged signaling molecules, that become active only when the designed protease cuts them.
Context
Designing proteins that bind a target has become routine in some labs. Designing enzymes that speed up a chemical reaction has been much harder, and success rates here, roughly one in five for the cysteine proteases, show it still is. Coverage also notes that many of the designs were much less efficient than natural enzymes.
The work is peer reviewed, but it is lab research. Cutting a disease-linked protein in a test tube or a cell culture is a long way from a treatment. See our guide to reading AI-for-science claims and to reading AI medicine news.
What to watch
- Animal studies testing whether a designed protease can reach and clear a target in the body.
- Whether other groups adopt the methods; an independent preprint from Westlake University also reports designed metalloproteases for amyloid-beta.
- Improvements in catalytic efficiency in later design rounds.
What we don’t know yet
- Whether the enzymes can be made active enough, delivered safely, and kept from triggering immune responses in the body.
- How specific the designs are across the full set of proteins in a cell, beyond the targets tested.
- Whether other labs can reproduce the success rates using the published methods.
Evidence and sources
- Peer-reviewed study De novo design of cysteine proteases
- Primary announcement Molecular 'scissors' open path to new therapies
- Independent reporting Researchers use AI to design protein-digesting enzymes
- Independent reporting Institute led by UW Nobel winner uses AI to build molecular scissors targeting disease proteins
- Independent reporting AI designs working molecular scissors that recognize and cut selected protein targets