In 2013, Jason McLellan and colleagues gave vaccine designers a treasure map by solving the structure of RSV’s prefusion F protein, the viral grappling hook before it springs shut. Dong and colleagues now send a new creature into that same forest: an mRNA vaccine candidate guided by AI-assisted antigen screening, as if the old structural biology compass has been upgraded with a very picky metal detector.

Here we observe respiratory syncytial virus, or RSV, in its natural habitat: the human airway, where it causes millions of infections and behaves with the social restraint of glitter at a craft table. It hits infants, older adults, and people with vulnerable lungs especially hard. The vaccine challenge has always been that RSV’s F protein is a shapeshifter. In its “prefusion” form, it exposes especially useful targets for neutralizing antibodies. Then it snaps into a postfusion form, and the immune system gets a less helpful mugshot.

That snap matters. A vaccine wants to show your immune system the right version of the intruder. Not “after the crime, wearing sunglasses,” but “right before entry, holding the crowbar.”

The Delicate Beast Called Prefusion F

The new paper, Highly efficient and durable Th1-biased protective immunity from a novel mRNA vaccine against RSV via AI-assisted antigen screening, reports an mRNA vaccine candidate called RSV-mF03. Like Moderna’s approved RSV vaccine mRESVIA, it uses mRNA to instruct cells to make RSV prefusion F protein. But the authors did not simply grab the old antigen and call it lunch.

They built a pool of engineered F protein variants, then used an AI screening model to predict which sequences might express well and provoke stronger immune responses. In plain English: they made a lineup of possible viral disguises, then asked a computational bloodhound which ones smelled most promising.

The winning candidate encoded a more stable prefusion F protein and was delivered using YK-009 lipid nanoparticles. In mice, RSV-mF03 reportedly produced durable neutralizing antibody titers more than 10 times higher than those induced by mRNA-1345, the vaccine marketed as mRESVIA. The authors also report complete protection after RSV challenge, with no evidence of vaccine-associated enhanced respiratory disease.

That last part is not decorative. RSV vaccine history includes painful lessons from older failed approaches, where immune responses could make later disease worse. A Th1-biased response is generally encouraging because Th2-skewed responses have been associated with enhanced respiratory disease concerns in RSV vaccine development. The immune system, like a nervous restaurant critic, does not just care what you serve. It cares how you plate it.

AI as the Quiet Tracker

The AI angle here is not “the machine invented a vaccine while wearing sunglasses.” It is more modest and more useful. Protein design has a brutally large search space. Change amino acids here, stabilize a loop there, preserve the right antibody target, keep expression high, avoid folding disasters. Biology loves making researchers fill out forms in triplicate.

AI-assisted screening can help rank candidates before expensive experiments begin. The animal work still matters. The cell assays still matter. The neutralization data still matter. AI is not the vaccine. It is the tracker moving ahead of the expedition, noticing bent grass and suspicious footprints.

This fits a broader trend. Recent work has used computational design to stabilize viral class I fusion proteins, including RSV F-like targets, while other groups have redesigned prefusion RSV F constructs to reduce unwanted immune responses or improve manufacturability. The field is no longer just asking, “Can we stabilize prefusion F?” It is asking, “Which stabilized version gives the immune system the cleanest, strongest, least weird lesson?”

Why This One Is Interesting

The intriguing claim is the combination: AI-assisted antigen selection, mRNA delivery, strong neutralizing titers, durability, complete protection in a challenge model, and no sign of enhanced disease in the reported experiments. That is a tidy little ecosystem. The antigen holds its shape, the lipid nanoparticle delivers the message, the immune system responds with neutralizing antibodies, and the Th1-skewed response avoids the historical swamp.

Still, mice are not tiny humans with worse email habits. Preclinical results often look majestic under laboratory lighting and then become more complicated in people. Human immune histories vary. RSV exposure differs by age. Older immune systems may respond differently. Safety needs large, careful trials. Manufacturing consistency matters. And a 10-fold antibody advantage in one experimental setting does not automatically translate into 10-fold better real-world protection.

But if these findings reproduce and survive clinical testing, the impact could be real. Better RSV vaccines could protect older adults, reduce hospitalizations, and eventually help guide vaccines for other unstable viral fusion proteins. The broader lesson may be bigger than RSV: combine structural biology, computational screening, and mRNA platforms, and vaccine design starts looking less like throwing darts in a dark room and more like tracking a shy animal through fresh snow.

The Field Around It

This study builds on a now-rich RSV vaccine landscape. McLellan’s 2013 Science work revealed the prefusion F target and helped make structure-based RSV vaccine design possible. Moderna’s mRNA-1345 later showed that an mRNA RSV vaccine could protect older adults in a phase 2/3 trial. In 2024, FDA approved mRESVIA for adults 60 and older, making it an important real-world benchmark. Meanwhile, newer studies continue refining prefusion F, including foldon-free trimers and general computational stabilization strategies.

In the quiet underbrush, the transformer does what transformers do: attends broadly, ranks patterns, and refuses to explain itself in a way that would satisfy a tired immunologist at 1 a.m. But paired with experiments, it may help researchers find better antigens faster.

For RSV, that could mean vaccines that show the immune system the virus at exactly the wrong moment for the virus and the right moment for us.

References

1. Dong K, Liu F, Zhang M, Zhou Y, Yu F, Wang W, Chai X, Li J, Song G. “Highly efficient and durable Th1-biased protective immunity from a novel mRNA vaccine against RSV via AI-assisted antigen screening.” Molecular Therapy (2026). DOI: 10.1016/j.ymthe.2026.01.031. PMID: 41612695

2. McLellan JS et al. “Structure of RSV fusion glycoprotein trimer bound to a prefusion-specific neutralizing antibody.” Science 340, 1113-1117 (2013). DOI: 10.1126/science.1234914

3. Walsh EE et al. “Efficacy and Safety of an mRNA-Based RSV PreF Vaccine in Older Adults.” New England Journal of Medicine 389, 2233-2244 (2023). DOI: 10.1056/NEJMoa2307079

4. Bakkers MJG et al. “A foldon-free prefusion F trimer vaccine for respiratory syncytial virus to reduce off-target immune responses.” Nature Microbiology 9, 3254-3267 (2024). DOI: 10.1038/s41564-024-01860-1

5. Gonzalez KJ et al. “A general computational design strategy for stabilizing viral class I fusion proteins.” Nature Communications 15, 1335 (2024). DOI: 10.1038/s41467-024-45480-z

6. Che Y et al. “Rational design of a highly immunogenic prefusion-stabilized F glycoprotein antigen for a respiratory syncytial virus vaccine.” Science Translational Medicine 15, eade6422 (2023). DOI: 10.1126/scitranslmed.ade6422

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.