Artificial intelligence–designed “smart” proteins can be programmed to turn on when they detect a certain target, researchers have shown. This feature allows them to work as biological switches from inside living cells, potentially acting as biosensors for medical, environmental, or biotech applications (Nat. Biotechnol. 2026, DOI: 10.1038/s41587-026-03081-9).
Proteins are molecular machines, and the way they can change in structure and function to switch between active and inactive states—a property called allostery—is responsible for essential biological functions such as metabolism and signaling.
In the body, “there are trillions of protein switches that form signal-processing networks operating in real time, [performing] very sophisticated computing operations that underlie all of our senses,” says Kirill Alexandrov, a biotechnology researcher at Queensland University of Technology, who led the study.
Bioengineers like Alexandrov want to connect proteins together to build computing or sensing systems. “But until very recently, we were limited only to naturally occurring proteins,” he says. “AI-driven protein engineering changes that significantly because now you can generate proteins never existed.”
The research team used AI-designed proteins to create modular structures that had receptor domains (which bind with a molecule to produce a signal) and reporter domains (which produce a signal, like color change or electrical current) to detect different substances, like small molecules, peptides, or nucleic acids. As proof that the switches can work inside cells, the researchers demonstrated electrochemical biosensors for steroid detection in Escherichia coli.
One of the obstacles scientists have faced in designing protein switches has been that the proteins would have to undergo big structural changes, and not many naturally occurring proteins do this. This study shows that when synthetic proteins are used, this need not be the case. In fact, subtle changes to how the protein moves was enough to turn activity on.
Adam Braunschweig, a biochemist at the City University of New York, who wasn’t involved in the research, calls it an elegant work. “This study demonstrates a design principle . . . to create sensors with a wide range of behaviors.” The different systems reported in the paper demonstrate that this strategy could be used to address a large number of problems in the biosensor space, he adds.
Alexandrov says that the researchers still have a lot of work to do to optimize these protein biosensors to be able to design “switches with this catalytic activity, or with this dynamic range.” Future work includes looking at parameters such as selectivity, specificity, maximum catalytic activity, latency, and more. “Optimizing them globally is a big challenge because it’s all sequence encoded, but the relationships between all of them are nonlinear.”
2026 American Chemical Society
