AI-designed miniproteins can now modulate G-protein-coupled receptors (GPCRs) in living cells, potentially unlocking access to hundreds of drug targets that have remained therapeutically out of reach. In a study published in Nature, researchers from Skape Bio and the University of Washington's Institute for Protein Design demonstrated that computationally designed proteins smaller than 100 amino acids can directly engage GPCR binding pockets and control receptor activity, addressing one of biology's most stubborn drug discovery challenges.

Why Have GPCRs Been So Hard to Target?

GPCRs represent the largest and most versatile family of cell surface receptors in the human body. Approximately one-third of all approved medicines act on these membrane proteins, treating conditions ranging from allergic rhinitis and pain to hypertension and schizophrenia. Yet despite their therapeutic importance, the human genome contains roughly 720 GPCR genes, and current drugs only target around 120 of them, leaving the vast majority of the "GPCRome" therapeutically unexplored.

The reason for this gap is not a lack of effort. GPCRs are notoriously difficult to work with. Many remain poorly characterized, with unknown natural ligands, unclear roles in the body, or expression patterns that complicate validation. Others have resisted traditional screening approaches because of low expression levels, poor solubility, or the absence of robust assays to test them.

Creating biologic drugs that actively modulate GPCR signaling has been particularly challenging. These receptors are complex, flexible membrane proteins that constantly shift between multiple active and inactive states, making it extremely difficult to design molecules that bind predictably. Additionally, their short extracellular loops limit accessibility and reduce the effectiveness of conventional screening and structural approaches.

"Even when designs are highly specific to a given receptor, tiny structural differences, sometimes on the scale of a single atom, can determine whether a receptor is active or inactive," explained Christoffer Norn.

Christoffer Norn, CEO and cofounder of Skape Bio

How Does the New AI-Designed Approach Work?

The research team tackled both the design and screening challenges simultaneously by combining computational protein design with native, cell-based screening. Rather than studying GPCRs in artificial laboratory conditions, they screened millions of AI-designed candidates directly in living cells, where the receptors maintain their natural conformation.

The workflow begins with selecting a target GPCR structure. The researchers choose whether they want an agonist (a molecule that activates the receptor) or an antagonist (one that blocks it), then start from the appropriate structural state. For agonists, they begin with an active-state structure and design proteins that lock the receptor into that active conformation. This structural bias dramatically improves the odds of success.

Once the structural backbones are established, AI tools design the protein sequences. The team uses AlphaFold-type prediction methods to identify which binders are most likely to function, then layers in additional biophysical constraints to further boost the hit rate. This process generates a library of up to 100,000 candidate designs.

The candidates then enter a high-throughput, cell-based screening system. Each cell produces a single candidate binder alongside the same GPCR target, using fluorescent proteins to create a direct visual readout. When a design binds its target, the receptor is retained in the endoplasmic reticulum (ER), and the cell appears yellow under the microscope. Designs that fail to bind allow the receptor to traffic to the cell surface, producing a green signal instead. This approach allows researchers to screen millions of cells and efficiently identify designs that genuinely work.

What Results Did the Study Achieve?

The research team generated functional miniproteins against 11 different GPCRs spanning multiple receptor classes implicated in itch and pain, cancer, metabolic disease, and migraine. In one striking example, a designed miniprotein antagonist matched the efficacy of an approved clinical drug in mobilizing hematopoietic stem and progenitor cells, while showing a reduced side-effect profile, including reduced leukocyte mobilization and no detectable systemic cytokine elevation.

The entire workflow from target selection to hit identification can take as little as three months for a new GPCR, a dramatic acceleration compared to traditional drug discovery timelines. This speed opens the possibility of systematically exploring the hundreds of GPCRs that have remained therapeutically inaccessible.

What Makes Miniproteins Uniquely Suited for This Challenge?

Miniproteins offer several advantages over conventional biologic drugs like antibodies and nanobodies. Their small size, fewer than 100 amino acids, allows them to slip into narrow binding pockets that are often inaccessible to larger molecules. This compact form factor enables high shape complementarity while maintaining strong and selective receptor engagement.

Beyond their structural advantages, miniproteins are highly customizable. Standard protein engineering strategies can be readily applied to tune their pharmacological properties and improve their drug-like characteristics:

• Half-life Extension: Fusion to an Fc domain extended circulation time from less than one hour to roughly 25 hours in mouse studies.
• PEGylation: Attaching polyethylene glycol chains can improve solubility and reduce immunogenicity.
• Albumin-Binding Motifs: Fusion to albumin-binding sequences can extend tissue exposure and circulation time.

"These are modular proteins. If you want a long half-life, you can engineer that in," noted Norn.

Christoffer Norn, CEO and cofounder of Skape Bio

How to Apply This Technology to Drug Discovery

• Target Selection: Identify GPCRs with known disease relevance but limited therapeutic options, prioritizing receptors that have resisted traditional drug discovery approaches.
• Structural Preparation: Obtain or model both active and inactive conformations of the target GPCR to guide whether agonist or antagonist designs are needed.
• Design Library Generation: Use computational design tools to generate 50,000 to 100,000 candidate miniprotein sequences, applying AlphaFold-type predictions and biophysical constraints to enrich for functional designs.
• Cell-Based Screening: Implement high-throughput microscopy-based screening in mammalian cells to identify candidates that modulate GPCR signaling in native cellular environments.
• Pharmacological Engineering: Apply standard protein engineering modifications to optimize half-life, solubility, and tissue exposure for the lead candidates.

What Does This Mean for the Future of Drug Discovery?

By combining de novo protein design with native, cell-based screening, researchers are beginning to transform one of the most complex target classes in biology into a more systematic and programmable space. The ability to rapidly generate and screen miniproteins against previously intractable GPCRs could unlock hundreds of new drug targets and accelerate the development of treatments for diseases that currently lack effective therapies.

The approach also demonstrates a broader principle in AI-driven drug discovery: computational design is most powerful when paired with functional validation in physiologically relevant contexts. Rather than relying solely on predictions, the researchers validated their designs in living cells, ensuring that the miniproteins work in the complex, dynamic environment where they would actually function as drugs. This combination of computational precision and biological reality may become a template for tackling other challenging drug discovery problems across the pharmaceutical industry.