MIT Researchers Use AI to Improve Nanoparticle Design for RNA Therapies

Researchers at MIT have developed a new method using artificial intelligence (AI) to design nanoparticles that can better deliver RNA vaccines and RNA-based therapies. Their study, published in Nature Nanotechnology, describes how they trained a machine-learning model to analyze thousands of lipid nanoparticle (LNP) formulations, allowing them to predict new and more effective RNA delivery materials.

RNA vaccines, such as those for SARS-CoV-2, rely on LNPs to protect mRNA and facilitate its entry into cells. Typically, LNPs consist of four components: cholesterol, helper lipid, ionizable lipid, and polyethylene glycol (PEG)-attached lipid. The many possible combinations of these components make it labor-intensive to test and find optimal formulations. To accelerate this process, the team created a library of about 3,000 LNP formulations and experimentally measured their RNA delivery efficiency. They used this data to train an AI model named COMET, which is inspired by transformer architectures used in language models like ChatGPT.

Unlike traditional AI drug discovery models that optimize single compounds, COMET predicts how multiple interacting components in LNPs affect their performance. After training, the model proposed LNP formulations that were tested in laboratory experiments, delivering mRNA encoding fluorescent protein to mouse skin cells. These AI-predicted particles performed better than existing ones, including some commercially available LNPs.

The team extended their approach to incorporate a fifth component, branched poly beta amino esters (PBAEs), polymers known for nucleic acid delivery. They generated around 300 LNP formulations with PBAEs to train the model again, enabling it to predict improved particles with these polymers.

Further, the researchers trained the model to predict LNPs that target specific cell types, such as Caco-2 colorectal cancer cells, successfully identifying efficient formulations in these contexts. They also used the model to find LNPs that can better withstand lyophilization, a freeze-drying step important for medicine storage.

Senior author Giovanni Traverso highlights the flexibility of this AI approach, stating, “This is a tool that allows us to adapt it to a whole different set of questions and help accelerate development.” The team is currently working on applying these nanoparticles for RNA therapies targeting diabetes and obesity, including GLP-1 mimics with effects similar to existing treatments like Ozempic.