AI designs protein switches that match a drug's effect in mice

The pharmaceutical century's defining paradigm — find molecules in nature or screen millions of compounds until something works — just yielded to a third. David Baker, who won the 2024 Nobel Prize in Chemistry for proving that proteins could be designed on a computer, published a paper in Nature last week that suggests the field has outgrown its own history.

Baker and colleagues at the University of Washington's Institute for Protein Design showed for the first time that AI can design miniproteins from scratch that actually control GPCR signaling — flipping these cell-surface receptors from on to off, rather than simply landing on them. GPCRs are everywhere: they govern vision, smell, adrenaline, insulin, nutrient sensing, and the action of roughly one in three approved drugs. They are also notoriously hard to drug, because their signaling switches sit in deep flexible pockets that most molecules cannot reach.

The Baker lab built miniproteins — compact proteins of fewer than 100 amino acids — that slip into those pockets and act as precision switches. Across 11 GPCR targets, including receptors involved in cancer, diabetes, obesity, migraine, itch, and pain, the team generated functional lead molecules that could either activate or block signaling depending on how they were designed. In a mouse study, one designed antagonist performed as well as a clinically used drug while showing fewer side effects.

The work also introduced a new screening system capable of testing up to 100,000 protein designs while keeping the receptors in their natural membrane environment — a departure from traditional screening methods that often require removing receptors from the cell or running at much lower throughput.

"Existing drugs such as antibodies bind to but often fail to activate or block GPCR signaling," said Edin Muratspahić, a postdoctoral scholar at the Institute for Protein Design and a first author of the study. "Seeing computationally designed miniproteins not only bind but actually control GPCR signaling in living cells was a defining moment for me."

The market stakes are large. GPCR-targeted medicines generate more than $200 billion in annual sales. GLP-1 agonists — Novo Nordisk's Ozempic and Wegovy — are the most visible recent example of what happens when a GPCR drug works. But those drugs were discovered through conventional means: finding what exists in biology and optimizing it. The Baker approach is different: design what does not exist and has never existed, atom by atom, on a computer.

Baker received the Nobel Prize in Chemistry in October 2024. His lab at UW has spent two decades building the computational tools for protein design from the ground up. This Nature paper — published May 21, 2026, the same week the award ceremony was still fresh in institutional memory — is the work of a laboratory operating at full ambition, not resting on a legacy.

A spinout called Skape Bio, founded in 2025 with roots in the Baker Lab and the BioInnovation Institute in Copenhagen, is already working to translate the platform into commercial drug candidates.

There are caveats. The 11 targets are leads, not drugs. Mouse studies are not human trials. Manufacturing miniproteins at pharmaceutical scale remains an engineering challenge. And the $200 billion market figure reflects the existing GPCR drug landscape, not revenue that will flow from this paper specifically. The path from designed miniprotein to approved medicine is long and littered with failures that looked promising in early animal studies.

But the conceptual shift stands. The pharmaceutical industry spent a century treating drug discovery as a problem of finding — finding molecules in nature, finding candidates through screening. The alternative was synthesis: making small modifications to known scaffolds. This paper describes something different. These molecules did not exist before the researchers imagined them. They were not found. They were built.

"The methods we are sharing in this new study form the roadmap for achieving all-computational design of protein ligands for any GPCR," said Christoffer Norn, a corresponding author and co-founder of Skape Bio.

If the approach scales — if the designed modulators prove safe and effective in humans — the implications extend beyond any single drug target. The most lucrative receptor family in pharmacology just became, for the first time, systematically designable. Not findable. Not screenable. Designable.