AI-rewritten: This is a summary of an article from Ars Technica, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Researchers at the University of Washington used artificial intelligence to design new enzymes capable of cutting specific proteins. While natural peptide bonds are extremely stable and can last hundreds of years without help, these AI-designed proteins were created to break those bonds efficiently. The team focused on two types of existing enzymes: a zinc protease from a crayfish and a cysteine protease from yeast. They used their own AI tools to replace the non-essential parts of these enzymes while keeping the core catalytic mechanisms intact.

The design process involved software called RFdiffusion2 to create protein backbones and ProteinMPNN to fill them with specific amino acids. The researchers ran this process dozens of times, generating hundreds of unique designs that did not match any known structures in existing databases. They tested these new enzymes by attaching them to fluorescent proteins to see if the enzymes could successfully cut the target sequence free.

The initial success rates were modest, with only 19 percent of cysteine protease designs working and a lower rate for the zinc version. However, by adding specific constraints like positioning water molecules correctly and swapping in alternative amino acids near the active site, the team significantly improved performance. Their best zinc protease design accelerated peptide bond cleavage by a factor of ten billion while remaining highly specific to its intended target.