AI tool shrinks proteins by 10–15% while preserving function

Duke University researchers, led by computational biologist Rohit Singh in collaboration with colleague Scott Soderling, have developed an AI-based program called Raygun that can strategically reduce a protein's size without compromising its function. The tool works by analyzing evolutionary patterns embedded in protein sequences to identify which amino acids are truly critical. As Singh explains, "This is an approach that essentially reasons from evolutionary patterns in protein sequences to figure out what the critical pieces are. You can ask for much more extensive substitutions than you would typically have by point-by-point mutations."

In initial tests, Raygun shrank two widely used fluorescent proteins—EGFP and mCherry—by 10–15% while preserving their luminescent properties; the resulting proteins still glowed when exposed to excitation. The findings were published in Nature (2026, DOI: 10.1038/s41586-026-10842-8).

The applications are significant for two major areas of biomedicine. In imaging, miniaturized fluorescent protein tags could reduce their intrusiveness when marking biological processes and tissues, lowering the risk that the tag itself distorts the system being observed. In gene therapy, one of the core challenges is the limited cargo capacity of delivery vehicles like adeno-associated viruses; shrinking therapeutic proteins could open up new treatment possibilities. Singh notes that the two leading strategies in protein engineering—de novo design (creating proteins from scratch) and Raygun's approach (optimizing existing sequences)—are not competitors. Instead, they complement each other: "De novo design is about creating something in an entirely new part of the protein space. Raygun is about exploring the space around it more effectively. So you could actually make a de novo design and then still apply Raygun on top of it to then explore the space around that specific novel design."

Looking forward, the team is developing an additional capability: the ability to expand proteins as well as shrink them. While counterintuitive, expanding a protein could be valuable if researchers want to integrate a new functional domain into an existing protein scaffold. This bidirectional flexibility suggests Raygun may become a versatile tool for protein optimization across a range of biotechnology applications.