Evolution took billions of years to produce the enormous variety of protein-cutting enzymes found in nature. A team of researchers has now skipped most of that schedule by keeping the one part that does the chemistry and asking AI to invent everything else.
The work uses AI-based protein design tools on two types of protease. In each case, the researchers retained the catalytic core and replaced the entire rest of the protein. The resulting enzymes can be tuned to slice specific proteins in half, and the team showed this by cutting proteins associated with neurological diseases.
The problem with a very stable bond
A protein is a chain of amino acids joined by peptide bonds. Each amino acid carries both an amino group and an acid group, so it can bond on two sides, which is what allows long chains to form. The bond itself is what you get when an amino group links to an acid and a water molecule is released. Amino acids also have side chains with their own distinctive chemistry, and these give a protein backbone its particular properties.
The bond is not exciting to look at, but it is tough. Left alone, peptide bonds are estimated to have a half-life of several hundred years. That is a long time for a molecule that, in practice, rarely gets to enjoy it.
Why nothing lasts
In the real world, proteins fall apart far faster than that. Organisms have evolved several ways to catalyze the breaking of peptide bonds, and those approaches have diversified into countless specialized protein-digesting enzymes. Proteases are everywhere, and they generally make short work of any protein that wanders into the environment.
That diversity is the product of a long evolutionary process, with each enzyme adapted to its own niche. It is impressive, and also not something a lab can order on demand.
Short-circuiting the catalog
The new approach treats a protease as two parts. One is the catalytic core, the bit that actually breaks the bond. The other is everything around it, which in nature determines how the enzyme behaves, and in particular what it targets.
The researchers kept the first part from two protease types and let AI design tools replace the second. The reported result is that these enzymes can be tuned to cut specific proteins in half, rather than being limited to whatever specificity evolution happened to hand them.
To show the designs work, the team pointed them at proteins associated with neurological diseases and cut them. That is a demonstration of function, and the report does not go further than that.
What we can and cannot say yet
The available account is brief, so several questions remain open:
- Which proteases. The two protease types used for the catalytic cores are not specified in the material available here.
- Which targets. The neurological disease proteins are described only in general terms.
- How well it works. No efficiency, specificity or other performance figures are given.
- What comes next. Whether the enzymes could serve any therapeutic purpose is not addressed.
So the honest summary is a proof of concept. Designers can now take the part of a protease that does the chemistry, discard the rest, and have AI write a new body around it that is steerable toward a chosen protein.
Nature spent a very long time on its protease catalog. The researchers appear to have decided that the interesting part was the one piece they could keep.