AI-guided formulation keeps experimental mRNA vaccines active after heat storage

What is new today

Nature Biotechnology published a peer-reviewed study on 28 September 2026 describing AGENT, short for Algorithm-Guided Experimental design for lipid Nanoparticle Thermostabilization. The work combines high-throughput laboratory screening with Bayesian optimization, a method that uses results from earlier experiments to choose the next combinations worth testing. The authors first posted the underlying manuscript on bioRxiv on 18 September. Today's development is therefore the reviewed journal publication and accompanying institutional account, not the first public appearance of the research.

The target is a practical weakness of mRNA medicines. Messenger RNA is fragile, and the lipid nanoparticles that protect and deliver it can also lose function during storage. Existing products use carefully controlled cold chains. The researchers sought solid, water-free formulations that would tolerate higher temperatures and could fit delivery systems such as dissolving microneedle patches. That could matter most where reliable refrigeration, transport and trained clinical staff are limited, although the study did not test a public-health rollout.[1][2][3]

How the AI-guided search worked

According to MIT's first-party account, the team began with nearly 50 excipients: sugars, salts and polymers that can help stabilize a formulation. Researchers measured how well each candidate protected lipid nanoparticles carrying mRNA for firefly luciferase. Cells emit light when that mRNA remains functional, giving the experiment a measurable proxy for bioactivity. Five promising excipients were then taken into the optimization loop. The algorithm proposed ratios, the laboratory tested two formulations at a time, and the results were returned to the model for the next proposal.

The paper reports six optimization iterations completed within one month, after manual exploration had taken months without finding a fully stable candidate. The team applied the approach to lipid compositions representative of the systems used in the Moderna and Pfizer-BioNTech Covid-19 vaccines. Those are clinically relevant analogues, not the commercial vaccine products themselves. The algorithm did not invent a vaccine antigen or diagnose patients; its role was to navigate a formulation space more efficiently than an exhaustive search.[1][2][3]

What the experiments found

The authors report that optimized solid-state formulations retained 100% bioactivity after storage at 37°C for more than two months. MIT also reports a formulation stored at room temperature for one year. In this context, bioactivity refers to the experimental ability of the stored mRNA nanoparticles to produce their encoded signal under the study's assay conditions. It is not the same as a regulator-approved expiry date, guaranteed field performance or proof that every dose remains unchanged.

The team then packaged SARS-CoV-2 mRNA antigens and evaluated immune responses. The peer-reviewed abstract says antigen-specific antibody and germinal-centre B-cell responses in rodents and non-human primates were non-inferior to freshly prepared soluble vaccines. MIT's account describes comparable responses in mice and reports that solid microneedle patches also generated responses similar to injections. One indexed methods excerpt identifies a mouse comparison using five animals and a 0.3-microgram mRNA dose; the accessible summaries do not expose a single denominator for every animal arm, so this report does not imply a larger sample than the paper documents.[1][2][3]

Why it matters — and where the evidence stops

A genuinely heat-tolerant mRNA platform could reduce freezer dependence, simplify emergency stockpiles and extend access to vaccines or RNA therapeutics in places with unreliable power and transport. A stable solid formulation could also make microneedle patches more practical, potentially reducing reliance on needles and specialist administration. Those are prospective benefits. The study does not measure distribution costs, vaccination coverage, wastage, patient outcomes or environmental savings, and it does not show that a finished product can survive all the temperature cycling, humidity and handling encountered in real supply chains.

The central limit is that this is preclinical evidence. Immune markers in mice and non-human primates do not establish safety, dosing, durability or protection in people. The experiments focus on SARS-CoV-2 antigens and two representative lipid compositions; the researchers' claim that an optimized carrier could accept other mRNA payloads still needs disease-specific validation. Manufacturing scale-up, sterile production, patch consistency and regulatory stability studies remain ahead. Full group sizes and all denominators should be read in the paper and supplementary methods rather than inferred from the institutional summary.

The authors report that part of the research was funded by the Gates Foundation. The preprint disclosure says senior author Ana Jaklenec has licensing, investment, consulting, advisory, speaking or sponsored-research relationships involving several organisations, including Moderna Therapeutics, Particles for Humanity, SiO2 Materials Science and VitaKey; the remaining authors declared no competing interests. These disclosures do not invalidate the results, but they strengthen the case for independent replication. The next meaningful evidence would be reproducible manufacturing data followed by human trials that compare safety, immune response and real storage performance with current formulations.[1][2][3]