AI-guided formulation stabilizes mRNA vaccines for months at body temperature

An AI-guided approach has identified solid-state formulations that keep messenger RNA (mRNA) vaccines active for months at 37 °C, potentially removing the ultracold storage barrier that limits global vaccine distribution. Researchers report today in Nature Biotechnology that their machine-learning framework, called AGENT, stabilized mRNA–lipid nanoparticles (LNPs)—the delivery vehicles used in approved mRNA vaccines—using formulations based on the lipid compositions of the Moderna and Pfizer-BioNTech COVID-19 vaccines.

The work was led by Ana Jaklenec at MIT’s David H. Koch Institute for Integrative Cancer Research, with first author Jinbi Tian. Instead of testing hundreds of thousands of possible excipient combinations manually, the team coupled high-throughput experiments with Bayesian optimization, an AI technique that selects the most informative next experiments from small datasets. AGENT converged on optimized formulations in six iterations over about one month, using roughly 48 samples for one LNP type and 12 for the other.

Those formulations were vacuum-dried into solid-state, water-free vaccines. The reconstituted particles retained full bioactivity after more than two months at 37 °C, and remained stable for at least a year at room temperature in extended testing. The stabilized vaccines also worked when incorporated into dissolving microneedle patches. In mice and cynomolgus macaques, solid-state vaccines—given by injection or patch—induced antibody responses non-inferior to those from freshly prepared soluble mRNA vaccines given by intramuscular injection. In mice, injected solid-state vaccines also elicited germinal center responses comparable to fresh soluble vaccines. The authors say this is the first demonstration of mRNA vaccine delivery by microneedle patches in non-human primates.

The findings address two practical barriers at once: temperature sensitivity and the need for trained personnel and sharps-safe delivery. Thermostable formulations could reduce cold-chain storage costs and vaccine wastage, though the authors frame those savings as estimates rather than measured outcomes.

Important caveats remain. The experiments were preclinical; the macaque study included small groups (n = 5), and human data are still needed. The team also observed bleb-like structures in reconstituted LNPs, a morphology not yet defined among standard critical quality attributes for approved LNP products. The authors say this will require standardized analytical methods and reproducible manufacturing before regulatory acceptance.

Still, because the method preserves clinically approved lipid compositions and uses widely available excipients, it offers a path that could be adapted to existing mRNA vaccine platforms. Further work will need to scale manufacturing and advance the patches toward clinical testing.

Reference: Tian, J., Tran, K.T.M., Pogostin, B.H. et al. Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI. Nature Biotechnology (2026). https://doi.org/10.1038/s41587-026-03331-w