Using artificial intelligence, MIT researchers have found a new way of designing nanoparticles that can more effectively provide RNA vaccines and other types of RNA therapies.
After training an automatic learning model to analyze existing thousands of delivery particles, researchers used it to predict new materials that would work even better. The model also allowed researchers to identify particles that would work well in different types of cells and discover ways to incorporate new types of materials into particles.
What we have done is 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 possible. “”
Giovanni Traveo, Associate Professor of Mitting Mecal Engineering, Gastroenterologist at Brigham and Women’s Hospital, and the main author of the study
This approach could considerably accelerate the development process of new RNA vaccines, as well as therapies that could be used to treat obesity, diabetes and other metabolic disorders, according to researchers.
Alvin Chan, a former MIT postdoc who is now a deputy professor at Nanyang Technological University, and Ameya Kirtane, a former MIT Postdoc who is now a deputy professor at the University of Minnesota, are the main authors of the new study, which appears today in Nanotechnology of nature.
Particle predictions
RNA vaccines, such as Sars-Cov-2 vaccines, are generally packed in lipid nanoparticles (LNP) for delivery. These particles protect the mRNA from decomposition in the body and help it enter into the cells once injected.
The creation of particles that manage this work more effectively could help researchers develop even more effective vaccines. Better delivery vehicles could also facilitate the development of mRNA therapies that code protein genes that could help treat a variety of diseases.
In 2024, the Traveo laboratory launched a multi-year research program, funded by the US Advanced Research Projects Agency for Health (ARPA-H), to develop new unmanageable devices that could make oral delivery of treatment and RNA vaccines.
“Part of what we are trying to do is developing means to produce more protein, for example, for therapeutic applications. Maximization of efficiency is important to be able to increase the quantity that we can produce cells, ”explains Traverso.
A typical LNP consists of four components – a cholesterol, an assistance lipid, an ionizable lipid and a lipid attached to glycol polyethylene (PEG). Different variants of each of these components can be exchanged to create a large number of possible combinations. Changing these formulations and testing everyone individually takes a long time, so Traverso, Chan and their colleagues decided to turn to artificial intelligence to help accelerate the process.
“Most models of AI in the discovery of drugs focus on the optimization of a single compound at a time, but this approach does not work for lipid nanoparticles, which are made of several components in interaction,” explains Chan. “To approach, we have developed a new model called Comet, inspired by the same transformer architecture that feeds large languages like Chatgpt. Just as these models understand how the words combine to form meaning, the comet learns how the different chemical components meet in a nanoparticle to influence its properties – like the way it can provide RNA in cells. »»
To generate training data for their automatic learning model, the researchers created a library of around 3,000 different LNP formulations. The team tested each of these 3,000 particles in the laboratory to see how effectively they could deliver their payload to cells, then fed all of this data in an automatic learning model.
After training the model, the researchers asked him to predict new formulations that would work better than existing LNP. They tested these predictions using the new formulations to supply the coding mRNA for a fluorescent protein with skin cells cultivated in a laboratory dish. They found that the LNP predicted by the model indeed operates better than the particles of the training data, and in certain cases better than the LNP formulations which are used commercially.
Accelerated development
Once the researchers have shown that the model could predict the particles that would effectively provide mRNA, they started asking additional questions. First of all, they wondered if they could train the model on the nanoparticles which incorporate a fifth component: a type of polymer called Esters amino poly beta branched (PBAE).
Traverso research and colleagues have shown that these polymers can effectively provide nucleic acids by themselves, so they wanted to explore if adding them to LNP could improve LNP performance. The MIT team created a set of around 300 LNP which also include these polymers, which they have used to form the model. The resulting model could then predict additional formulations with PBAE that would work better.
Then, the researchers decided to form the model to make predictions on the LNP which would work best in different types of cells, including a type of cell called Caco-2, which is derived from colorectal cancer cells. Again, the model was able to predict the LNP which would effectively provide mRNA to these cells.
Finally, the researchers used the model to predict which LNP could better resist lyophilization – a lyophilization process often used to extend the duration of conservation of drugs.
“This is a tool that allows us to adapt it to a set of different questions and to help accelerate development. We have made a large set of training that has entered the model, but you can then have much more targeted experiences and get useful outputs on very different types of questions, ”explains Traverso.
He and his colleagues are now working to integrate some of these particles into potential treatments for diabetes and obesity, which are two of the main targets of the project funded by Arpa-H. Therapies that could be delivered using this approach include imitations GLP-1 with effects similar to Ozempic.
This research was funded by the Go Nano Marble Center of the Koch Institute, the career of career development Karl Van Tassel, the MIT mechanical engineering department, Brigham and Women’s Hospital and Arpa-H.