MIT researchers have developed an artificial intelligence algorithm that rapidly optimizes lipid nanoparticle formulations, creating RNA vaccines that remain stable at room temperature for up to a year or near 100 degrees Fahrenheit for two months without requiring ultracold storage.
RNA vaccines proved their worth during the Covid-19 pandemic and are actively being developed to target other diseases, including cancer. Yet their greatest logistical roadblock has always been temperature. Because RNA is an exceptionally fragile molecule, it must be protected by lipid nanoparticles (LNPs) and kept in ultracold storage ranging from -20 to -80 degrees Celsius. That strict cold-chain requirement makes distributing the shots to regions lacking specialized refrigeration facilities a formidable challenge.
A team of engineers at the Massachusetts Institute of Technology has bypassed that limitation by enlisting machine learning. Using an artificial intelligence algorithm, the researchers adjusted the chemical formulation surrounding the delivery particles, yielding vaccines capable of withstanding extreme heat.
How Machine Learning Solved an LNP Screening Bottleneck
To make RNA vaccines heat-tolerant, scientists routinely add various excipients—such as sugars, salts, or polymers—to the protective lipid nanoparticles. While researchers in the lab of Ana Jaklenec and Robert Langer previously designed polymer-stabilized LNPs that handled higher temperatures, those particles differed from the FDA-approved formulations relied upon by commercial Covid-19 shots from Moderna and Pfizer. Adapting those standard FDA-approved recipes proved difficult.
“We were trying to use and screen excipients that we’ve previously used successfully to stabilize LNPs, but it just wasn’t working. It was really frustrating for the team.”
Ana Jaklenec, principal investigator in MIT’s Koch Institute for Integrative Cancer Research
To break through that roadblock, the team collaborated with researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). They deployed an algorithm designed to handle small datasets, marking a departure from typical automated experimental design. According to Mina Konaković Luković, an assistant professor of electrical engineering and computer science at CSAIL, the team had never applied their algorithms to a biological problem like vaccine stability.
“It was surprising to see how quickly the algorithm converged on a stable formulation – getting there in just a handful of iterations, rather than the exhaustive search that would normally be required.”
Mina Konaković Luković, assistant professor of electrical engineering and computer science at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL)
Testing Nearly 50 FDA-Approved Excipients in Cells
The research process involved testing nearly 50 FDA-approved excipients to evaluate how effectively each one stabilized RNA within an LNP. To measure performance, scientists used the particles to deliver mRNA encoding firefly luciferase—a protein that generates bioluminescence—into living cells. By quantifying the light emitted from the cells, the team calculated how well a given excipient protected the delicate mRNA cargo from degradation.
From those trials, the researchers selected five top-performing excipients. The AI algorithm then predicted optimal blending ratios to stabilize LNPs similar to those used in Moderna’s vaccines. By testing two formulations at a time in cells and feeding those results back into the system, the algorithm quickly narrowed in on stable mixtures that would otherwise have demanded thousands of manual experiments.
Performance and Immune Response in Preclinical Tests
The resulting formulations demonstrated remarkable resilience. When put to the test, the newly formulated vaccines remained stable at room temperature for an entire year. Even when subjected to nearly 100 degrees Fahrenheit for two months, the preparations held their structural integrity.
When these heat-resistant Covid-19 vaccines were administered to mice, they produced an immune response as strong as standard RNA vaccines modeled on commercial designs. Beyond simplifying standard refrigeration logistics, this stability opens the door to alternative delivery mechanisms, such as microneedle patches. These patches feature hundreds of tiny, vaccine-filled needles that dissolve upon application to the skin, releasing the vaccine without traditional needles.
Research Leadership and Publication Details
The findings were published in Nature Biotechnology. The work was led by graduate student Jinbi Tian and postdoc Khanh Tran as lead authors, with Ana Jaklenec and Robert Langer—the David H. Koch Institute Professor—serving as senior authors on the study.
By drastically cutting down the experimental workload required to reach stable formulations, the team’s machine-learning framework demonstrates that artificial intelligence can accelerate the design of biological delivery systems while overcoming major distribution barriers for next-generation vaccines.