Zhejiang University's AI-Powered Breakthrough in Membrane Protein Design

A multidisciplinary team at Zhejiang University (ZJU) has achieved a landmark advance in biological research by leveraging artificial intelligence (AI) to design novel 'armor' for membrane proteins. Published in the prestigious journal Nature on February 16, 2026, the paper titled "De novo design of GPCR exoframe modulators" introduces GPCR exoframe modulators (GEMs)—de novo engineered artificial transmembrane proteins that precisely target and modulate G protein-coupled receptors (GPCRs), a family of membrane proteins critical to human physiology.

GPCRs, which constitute the largest class of cell surface receptors, are embedded in the lipid bilayer of cell membranes and transduce extracellular signals into intracellular responses. With nearly 800 human GPCRs identified, they are involved in diverse processes like vision, taste, smell, neurotransmission, and hormone regulation. Approximately 40% of approved drugs target GPCRs, making them prime candidates for therapeutic intervention. However, their instability outside natural membrane environments has long hindered structural studies and drug development.

The ZJU breakthrough addresses this core challenge head-on. Led by corresponding authors Yan Zhang, Vice Dean of ZJU School of Medicine, and Min Zhang from the College of Computer Science and Technology, the team developed GEMs using a 'hallucination-like' AI design strategy inspired by natural transmembrane protein regulators. This interdisciplinary effort combines pharmacology, structural biology, and computational science, exemplifying ZJU's strength in AI-driven biomedical innovation.

Understanding the Challenge: Why Membrane Proteins Defy Traditional Design

Membrane proteins, including GPCRs, represent about 30% of the human proteome but account for over 50% of drug targets. Unlike soluble proteins, they require a hydrophobic lipid environment for stability, complicating expression, purification, and crystallization for techniques like cryo-electron microscopy (cryo-EM). Traditional approaches rely on orthosteric ligands binding the extracellular pocket, but allosteric modulation—targeting distant sites for subtler control—remains elusive due to design complexity.

Previous efforts used nanobodies or fusion proteins, but de novo design from scratch was unprecedented. ZJU's innovation shifts focus to the transmembrane domain (TMD), the seven-helix bundle core of class A GPCRs, enabling 'exoframe' binding that stabilizes without disrupting native function.

This TMD-targeting strategy not only stabilizes GPCRs but also allows biased signaling—favoring specific pathways like G protein over β-arrestin recruitment—for finer therapeutic tuning.

The AI Design Pipeline: From Hallucination to High-Affinity Binders

The ZJU team employed deep learning tools RFdiffusion for structure generation and ProteinMPNN for sequence optimization. They crafted GEMs with three structural prompts to bind specific TMD interfaces: TM1/2/4 (anchor), TM3/4/5 (biased allosteric modulator, BAM), and TM5/6/7 (negative allosteric modulator, NAM, or ago-PAM).

Thousands of candidates were generated in silico, filtered by AlphaFold predictions and binding affinity simulations. Four GEMs were selected for experimental validation on dopamine D1 receptor (D1R), a key neurotransmitter receptor implicated in Parkinson's and schizophrenia.

Key steps:

• Prompt engineering: Define GEM topology to cradle TMD helices.
• Hallucination generation: RFdiffusion hallucinates novel folds.
• Sequence design: ProteinMPNN assigns amino acids for stability.
• Filtering: Energy-based scoring and ColabFold validation.

This pipeline yielded GEMs with nanomolar affinity, far surpassing random designs.

Experimental Validation: Cryo-EM Structures and Functional Assays

Cryo-EM revealed GEMs binding precisely to TMD without occluding the orthosteric site. Resolutions reached 2.8–3.5 Å, deposited in PDB (e.g., 9LLE).

Functional assays in HEK293 cells showed:

• GEM_anchor stabilizes inactive D1R.
• GEM_BAM biases toward β-arrestin.
• GEM_NAM reduces agonist efficacy.
• GEM_ago-PAM boosts dopamine potency 10-fold, rescuing LoF mutants (e.g., R442H, common in Parkinson's).

These results confirm GEMs as versatile allosteric tools, opening doors to precision medicine.

Interdisciplinary Collaboration at ZJU: Medicine Meets Computing

The project exemplifies ZJU's cross-disciplinary ethos. Yan Zhang's pharmacology lab provided GPCR expertise, while Min Zhang's computational group drove AI innovation. Co-authors from Liangzhu Laboratory and Shanghai Institute for Advanced Study of ZJU highlight institutional synergy.

Yan Zhang noted: “This 'exoframe' approach provides new ideas for GPCR disorders like Parkinsonian syndromes lacking effective treatments.” Min Zhang emphasized AI's role in function-oriented membrane protein design.

ZJU, a top Chinese university, ranks high globally in AI and life sciences, fostering such breakthroughs through MOE Frontier Science Centers.

Implications for Drug Discovery and Biotechnology

GEMs enable high-throughput GPCR screening, mutant rescue, and biased signaling drugs. For D1R-related diseases, ago-PAMs could enhance dopamine therapy without side effects.

Broader: Scalable to 800+ GPCRs, accelerating ~$100B market. AI de novo design reduces trial-error, cutting costs from millions to weeks.

China's biotech surge, led by ZJU, positions it as AI-bio leader, rivaling US/Europe.

ZJU's Growing Leadership in AI-Driven Biology

ZJU invests heavily in AI-bio, with centers like Brain-Machine Integration. Recent feats include protein editing tools and genomic AI. This Nature paper, ZJU's first New Year stunner, underscores its trajectory.

Student Shizhuo Cheng, first author, exemplifies training blending computation and experiment.

Challenges Overcome and Future Directions

Challenges: AI hallucination accuracy, membrane expression, off-target binding. Solutions: Multi-prompt refinement, fusion constructs, extensive validation.

Future: Expand to other GPCRs, in vivo testing, clinical translation. Potential for synthetic biology, e.g., engineered signaling cascades.

Zhang Yan envisions GEMs for untreatable GPCR diseases.

Career Opportunities in AI and Biotech at Chinese Universities

This breakthrough highlights demand for AI-biotech experts. ZJU and peers seek faculty/postdocs in computational biology.Browse higher ed jobs, research positions.

Students: Pursue interdisciplinary PhDs; profs: Rate ZJU faculty on Rate My Professor. Career advice: Craft academic CVs.

China's higher ed ecosystem booms, with AI/bio hubs.

Global Impact and ZJU's Role in Advancing Science

ZJU's feat inspires worldwide, bridging AI and biology. As China invests in 'Double First-Class' universities, ZJU leads, producing Nature papers amid US-China tensions.

Outlook: GEM platform accelerates GPCR drugs, potentially saving lives in neurology, cardiology. For academics, it signals AI's transformative power.

Explore higher ed jobs, university positions, career advice.