MIT researchers have developed an AI-powered method to design nanoparticles that deliver RNA vaccines and therapies more effectively. By training a machine learning model on thousands of existing delivery particles, they were able to predict better-performing materials, tailor them for different cell types, and incorporate new components faster than ever before.
RNA vaccines, like those for COVID-19, rely on lipid nanoparticles (LNPs) to protect mRNA and help it enter cells. MIT’s approach uses a custom AI model called COMET, inspired by the transformer architecture behind large language models. Unlike typical drug-discovery AI that optimizes single compounds, COMET learns how multiple interacting ingredients in LNPs influence delivery efficiency.
To build the model, researchers created and tested around 3,000 LNP formulations. COMET’s predictions not only outperformed training examples but in some cases surpassed commercial LNPs. The AI also identified particle designs suited for specific cell types, such as colorectal cancer-derived Caco-2 cells, and even formulations that withstand freeze-drying for longer shelf life.
The system could significantly accelerate the development of RNA vaccines and treatments for conditions like obesity and diabetes by streamlining formulation testing and improving delivery efficiency. As lead researcher Giovanni Traverso noted, the tool can be adapted to answer new scientific questions and speed up breakthroughs in RNA-based medicine.