Forschern des MIT ist ein bedeutender Durchbruch gelungen. Sie nutzten die Möglichkeiten der künstlichen Intelligenz, um effizientere Nanopartikel für die Verabreichung von Impfstoffen und Therapien auf RNA-Basis herzustellen. Dieser Sprung nach vorn könnte die Entwicklung von Behandlungen für eine Vielzahl von Krankheiten vorantreiben, von Infektionskrankheiten bis hin zu Stoffwechselstörungen wie Diabetes und Fettleibigkeit.
RNA-Impfstoffe, wie sie zur Bekämpfung von COVID-19 eingesetzt werden, sind auf Lipid-Nanopartikel (LNP) angewiesen, um das genetische Material sicher in die Zellen zu bringen. Obwohl diese Nanopartikel eine wesentliche Rolle dabei spielen, dass die RNA ihr Ziel unversehrt erreicht, war die Herstellung der wirksamsten LNPs bisher ein langsamer, mühsamer Prozess. Das MIT-Team hat diesen Engpass jedoch mit Hilfe künstlicher Intelligenz umgangen.
By training a machine-learning model on a collection of thousands of previously tried LNP formulations, they engineered a system capable of predicting new and more efficient combinations – a system they named COMET. Not just content with improving efficiency, the model can even suggest formulations that are tailored to specific cell types and include novel materials to boost performance.
The team developed COMET, the machine-learning model, taking inspiration from the same transformer architecture that drives large language models like ChatGPT. COMET’s task was to understand how different chemical ingredients in a nanoparticle interact, dictating how efficiently it can deliver RNA into cells. As Alvin Chan, a former MIT postdoc and co-lead author of this groundbreaking study, explains, “COMET learns how these components come together to affect delivery efficiency.”
Für das Training von COMET wurden etwa 3.000 LNP-Formulierungen verwendet. Jede von ihnen wurde im Labor methodisch getestet, um ihre Effizienz bei der Übertragung von mRNA auf Zellen zu messen, so dass das Modell Muster erkennen und wirksamere Formulierungen vorhersagen kann. Bei Tests an im Labor gezüchteten Mäusehautzellen zeigten die von der KI vorhergesagten LNPs beeindruckende Ergebnisse und übertrafen viele bestehende Optionen, darunter auch einige, die derzeit kommerziell genutzt werden, was einen bedeutenden Moment in der Nutzung der KI zur Beschleunigung der biomedizinischen Forschung darstellt.
Having validated the model’s precision, the team undertook exploring more complex questions like whether the model could predict formulations that incorporate an additional fifth ingredient, such as branched poly beta amino esters (PBAEs). These polymers have shown promise in the delivery of nucleic acids on their own. In response, COMET was trained on an added set of around 300 LNPs containing PBAEs and successfully suggested new, more efficient combinations. This accomplishment again underscored the model’s versatility as it went on to predict LNPs optimized for specific cell types, including Caco-2 cells derived from colorectal cancer.
Another hurdle that the research team tackled was ensuring LNPs maintain stability during storage. COMET was used to predict which formulations could best withstand lyophilization – a freeze-drying technique used to prolong the shelf life of many medicines. The model identified stable candidates, demonstrating its utility in real-world applications.
The research forms part of a broader initiative, spearheaded by MIT and financially backed by the U.S. Advanced Research Projects Agency for Health (ARPA-H). The objective is to develop ingestible devices capable of administering RNA treatments orally, making them more accessible and easy to use. Giovanni Traverso, senior author of the study and associate professor of mechanical engineering at MIT, acknowledges that “Maximizing delivery efficiency is critical to producing enough therapeutic protein in the body.” He also applauds the fact “This AI-driven approach allows us to explore new formulations faster and more effectively than ever before.” Rising to the next challenge, the team is now incorporating these AI-designed nanoparticles into experimental treatments for obesity and diabetes.
This pioneering research was made possible through the generous funding from the GO Nano Marble Center at the Koch Institute, the Karl van Tassel Career Development Professorship, the MIT Department of Mechanical Engineering, Brigham and Women’s Hospital, and ARPA-H. Get more in-depth coverage of this innovation by reading the original article on MIT-Nachrichten.
Diese Website verwendet Cookies.