MIT team uses AI-guided optimisation to find heat-tolerant mRNA vaccine formulations

MIT researchers say an AI-guided search helped them find lipid nanoparticle formulations that let mRNA vaccines stay stable at higher temperatures, according to an MIT News release and a paper in Nature Biotechnology, both published on 28 September 2026. The results so far are from cells and animals, not people.

What the AI did

As the release and the paper's abstract describe it, the AI's job was to choose formulation ingredients and ratios, not to design the vaccine or its RNA. The paper describes a framework called AGENT (Algorithm-Guided Experimental design for lipid Nanoparticle Thermostabilization), which it calls "an artificial intelligence (AI)-driven framework that couples high-throughput experimentation with Bayesian optimization to identify thermostable mRNA−LNP formulations".

In plainer terms, it is an optimisation loop for choosing recipe ratios:

• The team first measured nearly 50 FDA-approved excipients (additives such as sugars, salts or polymers) one by one, by delivering mRNA for firefly luciferase into cells and measuring the light the cells gave off, the release says.
• They picked five of the most promising excipients and used the algorithm "to predict ratios of those excipients that would best stabilize LNPs similar to those used by Moderna".
• They tested two formulations at a time in cells, fed the results back and generated new predictions. After several rounds they chose one formulation to take into animal studies.

The paper's abstract says this took "six iterations completed within 1 month". The algorithm was built with MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), and the paper says the Bayesian optimisation is "implemented within the AutODEx framework", and gives a GitHub repository for a project named AutoOED as the location of the source code.

"The real beauty of this algorithm is that we can use it with small data sets," Ana Jaklenec, a principal investigator at MIT's Koch Institute, says in the release. Lead author Khanh Tran says: "Before we implemented the AI algorithm, we spent several months testing different combinations and also doing the prescreening of all the excipients that we could find, but nothing would get us to 100 percent stability."

What the results show, and under what conditions

• Form and storage. The particles, carrying Covid-19 mRNA antigens, were vacuum-dried into a solid, water-free form. The release says they were then stored at 37 degrees Celsius for two months, or at room temperature for one year. The abstract says the solid formulations "retained 100% bioactivity after storage at 37 °C for more than 2 months".
• Two nanoparticle types. The abstract says the work covered two lipid nanoparticles "representative of the Moderna (SM-102-based) and Pfizer-BioNTech (ALC-0315-based) compositions". These are lab formulations similar to those products, not the commercial vaccines themselves.
• Animals. The release says mice given the stored vaccines "showed equivalent immune responses" to mice given vaccines carried by nanoparticles similar to the original Moderna formulation. The abstract goes further on species and delivery: "In rodents and non-human primates, thermostable, solid-state vaccine formulations induced antigen-specific immune responses non-inferior to those elicited by intramuscular delivery of freshly prepared soluble vaccines."
• Patches. The team also made solid microneedle patches carrying a SARS-CoV-2 antigen, which the release says "generated an immune response similar to that produced by the injectable RNA vaccines".

The release and the abstract report immune responses; neither mentions protection against infection or any human testing. We could not read the full paper, which is behind a paywall, so we cannot give group sizes or the statistical detail behind "non-inferior".

The paper

"Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI" is a peer-reviewed article in Nature Biotechnology (DOI 10.1038/s41587-026-03331-w). The journal page lists it as received on 13 January 2026, accepted on 2 September 2026 and published on 28 September 2026. Graduate student Jinbi Tian and postdoc Khanh T. M. Tran are lead authors, and the paper says they "contributed equally"; Jaklenec and Robert Langer are senior authors.

The release says the work was funded in part by the Gates Foundation. The paper's funding section gives the Gates grant as INV-061250, and also names the Koch Institute Support (core) Grant P30-CA14051 from the National Cancer Institute, along with fellowship support for the lead authors. In the paper's competing-interests statement, covering "From Fiscal Year 2020 to the present", Jaklenec lists six entities, including Moderna Therapeutics, that she has received licensing fees from, invested in, consulted or sat on boards for, lectured for or conducted sponsored research with. Langer's equivalent list is given only as an external Dropbox link, which we did not open.

Why it matters

The release says RNA-LNP vaccines currently need to be kept at -20 to -80 degrees Celsius, "which makes it difficult to ship them to regions that don't have cold-storage facilities available", and that more heat-tolerant versions could be distributed more widely and could help with delivery methods such as microneedle patches. The researchers also say that once a heat-resistant formulation is found for a given nanoparticle, "it could be adapted to deliver any type of mRNA payload".

On our reading, the AI's part here is narrower than some headlines suggest: a small-data optimiser cut the number of lab rounds needed to tune additive ratios. Whether these formulations hold up in human trials and at manufacturing scale is not addressed in the release or the abstract.