Researchers Design Working Proteases From Scratch

Machine learning tools produced enzymes that cleave target peptides in both lab tests and mammalian cells.

Updated on Nov. 10, 2026 in Biotech

Isometric editorial illustration of a sculpted amber cluster of folds coiled around a cleft on a matte slate surface, evoking a newly designed enzyme.
Researchers used machine learning tools to design cysteine proteases from scratch, and 13 of 69 designs cleaved their target peptides in lab tests. AI Illustration. Upload story photo >

Choi and colleagues designed cysteine proteases de novo using machine learning-based tools, while Chen and colleagues designed metalloproteases with similar methods. Thirteen of 69 computationally designed cysteine proteases were catalytically active and cleaved their target peptide sequences.

Why it matters

De novo protease design is biochemically complex because of the catalytic sites these enzymes require. The results suggest computational methods can now produce functional enzymes with folds distinct from natural cysteine proteases.

Of 69 computational cysteine protease designs, 13 were catalytically active, and structural studies revealed an aspartate-histidine-cysteine catalytic triad. Whether the remaining designs failed for a common reason is not addressed in the extraction.

The players

Nature Biotechnology

Nature Biotechnology is a peer-reviewed scientific journal covering biotechnology research and applications.

The details

Choi and colleagues used RoseTTAFold Diffusion 2 for Molecular Interfaces (RFD2-MI), a version of the RoseTTAFold diffusion model optimized for designing enzyme active sites and protein-protein interactions. The designed enzymes adopted folds distinct from natural cysteine proteases and were enzymatically active in mammalian cells.

Timeline

  1. Nature Biotechnology published the de novo protease design article on November 10, 2026.

The Big Picture

The study extends the RoseTTAFold diffusion model, which was built for protein structure work, into designing functional enzyme active sites. Designing catalytic sites from scratch has long been far harder than designing stable protein folds.

Researchers gain a computational route to building enzymes for specific tasks rather than screening natural variants. The timeline for practical applications is not stated in the extraction.

The takeaway

Computational tools are beginning to design enzymes with defined catalytic chemistry, not just stable structures. Readers tracking biotech should watch whether these designs hold up outside controlled lab conditions.

Further reading

Explore more on Biotech developments shaping protein engineering.

Source note: This article includes information reported by Nature.