New AI model RiboNN boosts mRNA therapy development by predicting protein production efficiency across cell types

A groundbreaking artificial intelligence model, RiboNN, has been developed by researchers at the University of Texas at Austin and Sanofi to enhance the efficiency of mRNA-based treatments for viruses, cancers, and genetic disorders. This tool predicts how specific mRNA sequences will translate into proteins, offering a critical advancement in designing therapeutics that can precisely target disease mechanisms. The collaboration, which merged academic and industrial expertise, aims to reduce reliance on trial-and-error methods in drug discovery, accelerating the development of next-generation mRNA therapies. Messenger RNA (mRNA) serves as a blueprint for protein production, instructing cells to create essential molecules for health and disease. However, optimizing mRNA sequences to ensure sufficient protein output remains a complex challenge. RiboNN addresses this by analyzing subtle sequence differences that influence translation efficiency, the process by which cells convert mRNA into proteins. In tests across 140 human and mouse cell types, the model achieved twice the accuracy of previous methods, providing researchers with a powerful tool to tailor therapies for specific tissues, such as the liver or lungs. The project, led by Can Cenik, an associate professor of molecular biosciences at UT Austin, and Vikram Agarwal, head of mRNA platform design data science at Sanofi’s mRNA Center of Excellence, began as a curiosity-driven exploration. “We were curious whether cells coordinate which mRNAs they produce and how efficiently they are translated into proteins,” Cenik explained. This foundational research eventually led to RiboNN, which could revolutionize the design of mRNA therapies by enabling precise adjustments to sequences, maximizing protein output in targeted cells. RiboNN’s development relied on RiboBase, a comprehensive dataset compiled by UT undergraduate students who manually verified and completed data from over 10,000 experiments. This dataset, which measures translation efficiency across diverse cell types, was critical for training the AI model. The team, including UT graduate student Logan Persyn and Sanofi researchers Dinghai Zheng and Jun Wang, leveraged machine learning to create a system that identifies optimal mRNA sequences. The model’s potential extends beyond general protein production. In a companion paper published in Nature Biotechnology, the researchers discovered that mRNAs with related biological functions are translated at similar levels across different cell types, revealing a previously unknown coordination in protein synthesis. This insight could streamline the design of therapies targeting specific cellular processes. For instance, mRNA vaccines prompt cells to generate proteins that train the immune system to recognize pathogens, while therapies for genetic disorders might replace faulty proteins. RiboNN’s ability to predict translation efficiency could enhance these applications, ensuring therapies are both effective and precisely delivered. “This opens an opportunity to modify mRNA sequences to boost protein production in specific cell types,” Cenik noted, emphasizing the tool’s role in advancing personalized medicine. The partnership between UT Austin and Sanofi was facilitated by the university’s Discovery to Impact office, which helped establish a research agreement. The collaboration highlights the growing synergy between academia and industry in tackling AI-driven challenges in biotechnology. With this innovation, researchers can now design mRNA therapies more strategically, reducing the time and cost associated with traditional experimentation. As the AI race intensifies, tools like RiboNN underscore the importance of data-driven approaches in unlocking the full potential of mRNA technology for treating a wide range of diseases.