AI designs working molecular scissors that recognize and cut selected protein targets

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility: Proteases are naturally occurring enzymes that break down proteins into peptides and even single amino acids. They are useful in medicine and industrial processes, where they clean tissue, digest waste and help in the manufacture of everyday products.

Now scientists have used artificial intelligence to design new versions from scratch. AI tools have already successfully generated new protein shapes and structures, but using AI to create functional proteases is a significant leap forward. That's because the resulting enzyme has to both recognize a chosen protein sequence and cut it efficiently.

To see if it could be done, David Baker at the University of Washington and his colleagues combined several advanced artificial intelligence tools to create both cysteine and zinc proteases. First, they used a version of RFdiffusion2 called RFD2-MI to design the 3D shape and core backbones of the proteins. Next, they applied ProteinMPNN and a related program called EnhancedMPNN to select the amino acid sequences needed to build the enzymes and their targets, helping them fit together correctly.

To find options that actually worked, the researchers repeated this process dozens of times for both types of protease. They then used AlphaFold 3 to predict the final shape of these digital versions of proteins and check how closely they matched the intended designs. With that done, it was time to test the designs.

That involved using bacteria to produce a single, long protein chain that linked the candidate enzyme to its intended target sequence and a glowing marker protein. The team tested 69 versions of the cysteine protease and 135 of the zinc protease. An enzyme design would be deemed workable if it chopped the target sequence and set the marker free.

Their findings on cysteine proteases and zinc proteases are detailed in two papers published in Nature. Out of the 69 cysteine protease candidates, 13 successfully cut their targets. The zinc protease designs performed even better, with 36% of the 135 candidates cutting their matching peptide targets.