Researchers Use AI to Design Proteases That Cut Up Other Proteins

Researchers have used AI-based protein design tools to build enzymes that cut specific proteins in half, taking the catalytic core of two protease families and replacing the entire rest of each protein with designed structure. The team demonstrated the resulting enzymes on proteins associated with neurological diseases, according to a report published by Ars Technica in October 2026.

The work matters because it compresses a process that evolution stretched over immense timescales into a design exercise. Proteases — enzymes that break peptide bonds — diversified across billions of years into countless specialized protein-digesting forms. The researchers skipped that diversification entirely and produced tunable molecular scissors on demand.

Why are peptide bonds so hard to break?

A protein is a chain of amino acids held together by peptide bonds. On their own, those bonds are extremely stable. Researchers estimate a half-life of several hundred years.

Yet most proteins in Earth's environment do not survive anywhere near that long. Organisms evolved a handful of mechanisms to catalyze the break-up of peptide bonds, and those mechanisms diversified into the specialized protein-digesting enzymes found throughout nature. Proteases are, as Ars Technica puts it, everywhere, and they make short work of proteins that enter the environment.

The chemistry itself is simple. A peptide bond forms when an amino group links to an acid group, releasing a water molecule. Amino acids carry both groups, so each one can bond on two sides. That property builds the long chains found in proteins, and the distinctive side-chain chemistry of each amino acid gives every backbone its specific properties.

What did the AI design approach change?

The researchers took the catalytic core from two types of protease and replaced the rest of the protein around it with AI-designed structure. The result: enzymes that could be tuned to cleave chosen target proteins rather than whatever a natural protease happens to digest.

This is the short-circuit. Evolution produces new protease specificities through slow divergence of whole proteins. The AI pipeline starts from a functional core and designs a new body around it, aiming the cutting activity at a target the researchers select.

The proof of concept carried medical weight. The team directed its designed enzymes at proteins associated with neurological diseases and cut them, which points toward a possible route for therapeutic strategies that degrade disease-linked proteins at specified points.

Why does this matter for AI-driven biology?

The study adds to the case that AI protein design has moved beyond folding prediction into building functional enzymes to order. Designing a protease that cleaves a chosen substrate is a harder test than designing a binder, because the enzyme must recognize the target and catalyze a chemical reaction at a specific site.

The demonstrated cuts in neurological-disease-associated proteins suggest the method could support targeted protein degradation research, an area of intense pharmaceutical interest. Whether the designed enzymes hold up in living systems remains the next question for the field.

The full findings are available in the Ars Technica report of October 2026.