AI framework AGENT yields mRNA vaccines stable at 37 °C for two months

Researchers led by J. Tian published a study in Nature Biotechnology on 28 September 2026 describing AGENT, an AI-driven framework that combines high-throughput experimentation with Bayesian optimisation to engineer thermostable, solid-state mRNA vaccines ¹².

What the study reports

The formulations produced by AGENT retained full bioactivity after storage at 37 °C for 2 months ². They elicited immune responses that were non-inferior to fresh vaccines, and they supported cold-chain-free delivery using microneedle patches in rodents and nonhuman primates ². The published summary describes the improvement as one of thermostability in RNA vaccines achieved with AI, and places the work across biomaterials, biomedical engineering, nanoparticles and SARS-CoV-2 vaccine safety and efficacy studies ¹².

The approach builds on earlier work in the same journal on a microneedle vaccine printer for thermostable COVID-19 mRNA vaccines, published in Nature Biotechnology in 2024, and on a 2026 report in Drug Delivery and Translational Research of an mRNA-loaded microneedle patch using a clinically relevant lipid nanoparticle composition ². The optimisation method itself is long established: the authors cite a 2015 review of Bayesian optimisation in Proceedings of the IEEE ².

Two things are being claimed at once. The first is a product property, a solid-state mRNA–lipid nanoparticle vaccine that survives two months at body temperature without loss of bioactivity ². The second is a process property, that a data-efficient AI loop paired with high-throughput screening can reach such a formulation rapidly ¹². The published summary characterises AGENT as enabling "rapid engineering" of these formulations, without quantifying how many experiments or how much calendar time the loop required ².

What it means for the market

For vaccine manufacturers and the teams that plan their distribution, the relevant variable is storage at 37 °C rather than refrigeration or ultra-cold handling, combined with microneedle patch administration ². The study cites a 2025 review in Outbreak Management and Response on strategies for efficient storage and distribution of vaccines in resource-constrained environments, which is the setting the thermostability claim speaks to ². The sources do not quantify any cost, logistics or coverage benefit, so the commercial value of removing the cold chain is not established by this work.

For formulation and CMC groups, the interest is in the method. AGENT is presented as data-efficient, which matters where each experimental cycle is expensive and screening budgets are fixed ¹². The sources give no comparison against conventional formulation development in time, cost or number of candidates tested, so buyers of AI platforms have no benchmark here to hold a vendor to.

Nothing in the sources addresses regulation, reimbursement or procurement. No regulatory engagement, clinical trial, licensing arrangement or manufacturing partner is named ¹². Whether solid-state mRNA formulations and patch delivery can meet comparability expectations for an already approved product class remains unaddressed by this publication.

Caveats

The evidence is preclinical. Immunogenicity was assessed in rodents and nonhuman primates, with no human data reported, and the sources give no animal numbers, no study design details and no immunological endpoints or effect sizes ². Non-inferiority is asserted against fresh vaccine without a stated margin ². Stability is reported at a single temperature for a single duration, 37 °C for 2 months ².

The detail available publicly is also limited: the Nature Biotechnology research briefing is a summary of the full paper, and both the briefing and the primary article sit behind subscription access, with individual article purchase listed at USD 39.95 ².