An innovative artificial intelligence approach enhances medication and vaccine discovery by forecasting the efficacy of certain mRNA sequences in protein synthesis across different cell types. The recent advancement, created through a collaboration between The University of Texas at Austin and Sanofi, facilitates the prediction of protein production in cells, thereby reducing the necessity for trial-and-error research and expediting the development of the next generation of mRNA treatments.

Messenger RNA (mRNA) encodes directives for protein synthesis, facilitating bodily growth and the execution of daily physiological functions. The development of novel mRNA vaccines and therapeutics to combat viruses, cancers, and genetic disorders represents a highly promising domain in health and medicine. This endeavour often entails the complex task of stimulating patient cells to generate sufficient protein from therapeutic mRNA to effectively address disease.

The novel model, termed RiboNN, aims to facilitate the discovery of innovative mRNA-based therapies by elucidating factors that maximise protein production or enhance targeting of certain organs, such as the heart or liver. The team presented their model today in one of two associated publications published in the journal Nature Biotechnology.

“At the inception of this project over six years ago, no apparent application existed,” stated Can Cenik, an associate professor of molecular biosciences at UT Austin, who co-directed the initiative alongside Vikram Agarwal, head of mRNA platform design data science at Sanofi’s mRNA Centre of Excellence. We sought to determine if cells synchronise the production of mRNAs and the efficiency of their translation into proteins. This represents the significance of curiosity-driven research. It establishes the groundwork for innovations such as RiboNN, which only materialise subsequently.

The project was facilitated by financial assistance from the National Institutes of Health, The Welch Foundation, and the Lonestar6 supercomputer at the Texas Advanced Computing Centre of the University of Texas.

In experiments involving over 140 human and animal cell types, RiboNN demonstrated approximately double the accuracy in predicting translation efficiency compared to previous methodologies. This advancement may enable researchers to forecast cellular behaviours, perhaps accelerating treatments for cancer, infectious disorders, and hereditary conditions.

The process by which cells in your body synthesise proteins can be likened to a team of chefs preparing cakes. To synthesise proteins, the ribosomes within your cells reference the instructions in your unique DNA, transcribe these instructions onto messenger RNAs (mRNAs), and subsequently assemble amino acids according to the specified sequences to produce proteins.

An mRNA vaccination or treatment prompts cellular mechanisms to synthesise proteins. A vaccination may generate a protein present on the surface of a harmful virus or cancer cells, thereby signalling the immune system to develop antibodies against the virus or cancer. In instances of disorders resulting from genetic mutations, they may generate a protein that the body is unable to synthesise enough, so ameliorating the illness.

Prior to constructing their novel predictive model, Cenik and the UT team meticulously assembled a dataset of publically accessible information from over 10,000 trials that assessed the efficiency of mRNA translation into proteins across various human and mouse cell types. After the creation of the training dataset, AI and machine learning specialists from UT and Sanofi collaborated to develop RiboNN.

Cenik stated that a primary objective of the predictive tool is to eventually develop therapies specifically aimed at a distinct cell type. He is also an affiliate faculty member at UT’s Oden Institute for Computational Engineering and Sciences and a CPRIT scholar, receiving research funding from the Cancer Prevention and Research Institute of Texas.

“Perhaps a next-generation therapy is required to be produced in the liver, lung, or immune cells,” he stated. This presents an opportunity to modify the mRNA sequence to enhance the synthesis of that protein in the specific cell type.

In a subsequent article published in Nature Biotechnology, the researchers illustrated that mRNAs with analogous biological functions are translated into proteins at comparable amounts across various cell types. Researchers have long recognised that the transcription of functionally related genes into mRNAs is coordinated; however, it had not been previously demonstrated that the translation of mRNAs into proteins is similarly coordinated.

Undergraduate student researchers at UT meticulously verified the existing data for accuracy and supplemented missing information to construct RiboBase, the dataset required for training the AI model. The teams who worked on the development of RiboNN comprised Logan Persyn, a PhD student in computer science at UT, together with Dinghai Zheng and Jun Wang from Sanofi. The Discovery to Impact office at UT facilitated the collaboration between UT and Sanofi by formulating a research agreement.

Original Publication
Journal: Nature Biotechnology
DOI: 10.1038/s41587-025-02712-x
Method of Research: Experimental study
Subject of Research: Cells
Article Title: Predicting the translation efficiency of messenger RNA in mammalian cells
Article Publication Date: 25-Jul-2025