MIT AI Speeds RNA Vaccine & Therapy Development
MIT engineers have unveiled a groundbreaking application of artificial intelligence, developing a machine-learning model capable of designing nanoparticles that deliver RNA to cells with unprecedented efficiency. This advance promises to accelerate the development of not only RNA vaccines, akin to those used against SARS-CoV-2, but also a new generation of RNA-based therapies for conditions ranging from obesity to diabetes and other metabolic disorders.
RNA vaccines and many emerging RNA therapies rely on sophisticated delivery systems, typically tiny fat-based spheres known as lipid nanoparticles (LNPs). These nanoparticles serve a critical dual role: protecting the delicate messenger RNA (mRNA) from degradation within the body and facilitating its entry into target cells upon injection. Enhancing the efficiency of these delivery vehicles is paramount for creating more potent vaccines and for expanding the therapeutic potential of mRNA, which carries genetic instructions for producing beneficial proteins in the body.
Traditionally, optimizing LNP formulations has been a painstaking, time-consuming process. A typical LNP is composed of four key ingredients—a cholesterol, a helper lipid, an ionizable lipid, and a lipid attached to polyethylene glycol (PEG). The sheer number of possible combinations, arising from swapping different variants of each component, makes exhaustive experimental testing impractical. Recognizing this bottleneck, Giovanni Traverso, an associate professor of mechanical engineering at MIT and senior author of the study, along with lead authors Alvin Chan and Ameya Kirtane, turned to artificial intelligence.
Their innovative solution is a new model dubbed COMET, drawing inspiration from the “transformer” architecture that underpins large language models like ChatGPT. While most AI models in drug discovery focus on optimizing single compounds, COMET is uniquely designed to understand how multiple chemical components interact within a nanoparticle to influence its properties, such as its ability to deliver RNA effectively into cells. “What we did was apply machine-learning tools to help accelerate the identification of optimal ingredient mixtures in lipid nanoparticles to help target a different cell type or help incorporate different materials, much faster than previously was possible,” Traverso explained.
To train COMET, the researchers meticulously assembled a library of approximately 3,000 distinct LNP formulations. Each of these particles was laboratory-tested to quantify its RNA delivery efficiency, with the resulting data then fed into the machine-learning model. Once trained, COMET was tasked with predicting novel formulations that would surpass existing LNPs. Experimental validation confirmed the model’s prowess: the AI-predicted LNPs demonstrated superior performance in delivering mRNA encoding a fluorescent protein to mouse skin cells in lab dishes, in some cases even outperforming commercially available LNP formulations.
The researchers further explored COMET’s versatility. They successfully trained the model to incorporate a fifth component—branched poly beta amino esters (PBAEs), a type of polymer known for its ability to deliver nucleic acids—into LNPs, leading to predictions for even better-performing hybrid particles. The model also proved adept at predicting LNPs optimized for delivery to specific cell types, including Caco-2 cells derived from colorectal cancer. Furthermore, COMET could forecast which LNP formulations would best withstand lyophilization, a freeze-drying process crucial for extending the shelf-life of many medicines.
This AI-driven methodology, published in Nature Nanotechnology, represents a significant leap forward in drug discovery, offering a flexible tool that can be adapted to a wide array of research questions. The team is now actively working to integrate some of these AI-designed particles into potential treatments for diabetes and obesity, including GLP-1 mimics similar to drugs like Ozempic. This effort is part of a multi-year research program funded by the U.S. Advanced Research Projects Agency for Health (ARPA-H), specifically targeting the oral delivery of RNA treatments and vaccines. By maximizing the efficiency of protein production within cells, this approach holds immense promise for developing highly effective new therapies, dramatically shortening the development timeline from concept to clinical application.