AI-designed metalloproteases selectively cut disease-associated proteins in laboratory tests
Published in Artificial Intelligence, Bioinformatics, Biology, Drugs development & discovery, Journals.
Published October 15, 2026.
A study published in Nature on October 14 reports computationally designed metalloproteases: enzymes that use zinc to help cut peptide bonds. Anqi Chen, Kejia Wu and colleagues used RFdiffusion methods to position catalytic sites beside pockets that recognize peptide substrates.
Of 135 designs tested, 36% cleaved their intended peptide substrate at the targeted bond. The researchers also designed enzymes that selectively cut disease-associated proteins, including TDP-43, amyloid-beta and serum amyloid A.
The work extended to engineered cytokines and receptor antagonists that could be activated by the designed proteases, illustrating a way to control biological signals with matched molecular components.
These are laboratory demonstrations of enzyme design, not evidence that the proteins can safely treat disease in people. Therapeutic use would require further testing of delivery, unwanted cleavage, immune responses and effectiveness in relevant biological settings. The advance is the ability to design selective protein-cutting tools; clinical applications remain prospective.
Source: Chen and colleagues, “Computational design of metalloproteases,” Nature, October 14, 2026.