An MIT-led study published on 28 September 2026 used AI to optimise additives around RNA-carrying lipid nanoparticles. Its AGENT workflow reached thermostable solid formulations through six experimental rounds within a month. The challenge is keeping RNA and its delivery particles functional through drying, warm storage and rehydration.[1][2]
Two mouse experiments show what survived storage. Reporter RNA instructs cells to make an easily detected protein. Vaccine RNA in the second experiment encoded the receptor-binding domain of SARS-CoV-2.[2]
Two storage experiments, two endpoints
| Stored material | Exposure | Readout |
|---|---|---|
| Reporter RNA in formulation B | One year at room temperature | Protein-expression signal six hours after injectionSource 2 |
| SARS-CoV-2 RBD RNA in formulation A | One month at room temperature or 37°C | Antibodies measured at week fiveSource 2 |
Choosing experiments worth doing
The optimisation used the AutODEx framework.[1]
AutODEx models a laboratory task as a mapping from adjustable inputs to measured outcomes. It builds a cheap statistical approximation of that mapping, then proposes experiments that either look promising or help resolve uncertainty.[3]
That balance matters when every answer consumes materials and laboratory time. Testing only near the best formulation found so far can miss a better region. Testing unfamiliar combinations indefinitely can waste effort. Bayesian optimisation uses both predicted performance and uncertainty to choose the next batch.[3]
Consider a simplified search with five independently adjustable ingredients. Five possible levels for each create 3,125 combinations; ten levels create 100,000. This arithmetic illustration ignores practical constraints and is not the study’s experimental count. It shows why making the grid finer quickly becomes expensive.
The platform accepts measurements entered by researchers or returned by connected evaluation software. Those measurements update the model before another selection. Automation concerns the choice and organisation of experiments; the biological assay still provides the evidence.[3]
A storage window that people can use
Heat stability matters along an entire delivery journey. WHO identifies refrigeration capacity, ice-pack production and transport logistics as constraints on vaccine access. A formulation that tolerates a longer warm interval could therefore change which journeys are practical, if that tolerance is established for the finished vaccine.[4]
WHO’s framework for extended controlled temperature conditions requires approved maximum temperatures and durations in the package information. The useful destination is a specific, dependable storage window: how warm, for how long, with what performance preserved.[4]
Sources & further reading
- Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI Nature Biotechnology · 28 Sep 2026 · Accessed 12 Oct 2026
- Study supplementary information Nature Biotechnology · 28 Sep 2026 · Accessed 12 Oct 2026
- AutODEx: automated experimental design MIT CSAIL · Accessed 12 Oct 2026
- Extended Controlled Temperature Conditions World Health Organization · Accessed 12 Oct 2026