A team at MIT has used an AI algorithm to design a new RNA vaccine formulation that keeps the lipid nanoparticles (LNPs) carrying the RNA from falling apart at high temperatures. According to an MIT news release dated September 28, the new formulation can be stored for a year at room temperature and for about two months at roughly 37°C (98°F). The findings are published in Nature Biotechnology.
Most existing mRNA vaccines require frozen storage between -20°C and -80°C, which makes them dependent on a cold chain for shipping and warehousing. That has been a direct reason COVID-19 vaccines rolled out slowly in low-income countries, and it's also the backdrop for the Gates Foundation funding behind this research.
What the algorithm did
The research was led by Ana Jaklenec and Professor Robert Langer, principal investigators at MIT's Koch Institute for Integrative Cancer Research, with graduate student Jinbi Tian and postdoc Khanh Tran also taking part. The algorithm came from Mina Konaković Luković, an assistant professor at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL).
The process ran in stages. The team first measured how nearly 50 FDA-approved ingredients individually stabilized RNA, then narrowed the list to the five best performers. The hard part was figuring out the right ratio to mix them in — the combination space was too large to test by hand, which would have taken months. The algorithm predicted which ratios were most likely to work next based on a small set of existing results; the lab ran those experiments and fed the results back in, converging over several rounds. The whole process took just a few weeks.
"The real beauty of this algorithm is that we can use it with small data sets."
— Ana Jaklenec
Luković said the algorithm converged on a stable formula faster than the team expected.
Why small data matters here
Biological experiments are expensive to generate data from — each round of formula testing involves synthesis, storage, and testing that can take days, making it hard to accumulate the scale deep learning normally needs. Most of the AI-in-the-lab stories over the past year or two have centered on fields with large public databases behind them, like protein structure or molecule generation. Formulation work, where every lab's dataset is small, is better suited to "active learning" style small-sample optimization: the model only suggests which experiment to run next, and real measurements do the judging.
Choosing five ingredients already on the FDA-approved list was also a practical decision. Switching to an entirely new chemical would mean starting safety review from scratch; reusing approved excipients gives a shorter path toward clinical trials.
How far from the clinic
The immunogenicity data so far comes from mice, and the research team says the immune response is comparable to Moderna's original formula. The release doesn't address how it performs in humans, batch-to-batch stability at large-scale production, or whether regulators will accept a room-temperature storage label — and no company has announced plans to take it into clinical trials yet.
In China specifically, the comparison is concrete. Domestic mRNA vaccines already on the market or in development also depend on a cold chain, and distribution costs to western and remote regions have long been a sticking point. If a room-temperature-stable LNP formulation holds up in human trials, the biggest impact would fall on vaccine logistics and grassroots vaccination rather than on the labs themselves. The method used in the paper isn't especially complex, so domestic delivery-system teams could apply the same approach to screen their own excipients without a high barrier to entry.
Sources: CocoLoop, MIT News, Nature Biotechnology paper details; verified for storage temperature and duration, number of candidate ingredients, researcher credits, and funding source.