RiNALMo: an AI model that deciphers the language of RNA to power next-generation therapeutics

RNA plays a vital role in how our genes are expressed and how diseases develop. Yet, because RNA molecules constantly change shape, understanding how they work has long been a major scientific challenge.

In a leap for biomedical science, researchers from the A*STAR Genome Institute of Singapore (A*STAR GIS) and A*STAR Bioinformatics Institute (A*STAR BII), in collaboration with the University of Zagreb, have developed RiNALMo, an artificial intelligence (AI) model that can “read" and interpret RNA sequences like a language.  This breakthrough helps scientists predict how RNA behaves in the body, potentially accelerating the development of RNA-targeted and RNA-based treatments.

Why RiNALMo Matters

RNA lies at the centre of many modern medical innovations, from mRNA vaccines to gene therapies. However, designing RNA drugs or predicting their behaviour has long been slow and complex. RiNALMo addresses this challenge by offering a powerful new way to understand and engineer RNA for therapeutic use.

By learning the “grammar” of RNA from over 36 million sequences, RiNALMo can predict how RNA folds, interacts, and functions. These are capabilities that could speed up drug discovery by reducing the need for lengthy lab experiments, as well as improve vaccine and RNA therapy design, enabling more precise and stable molecules. Additionally, it supports diagnostic innovation, by revealing how RNA changes in disease conditions.

These advances could accelerate the development of a new generation of RNA-based medicines, bringing more precise and effective treatments to patients faster.

How RiNALMo Works

RiNALMo is the world’s largest RNA language model. It was trained on 36 million RNA sequences, giving it exposure to a vast diversity of biological data, and powered by 650 million parameters, that enable it to detect subtle patterns smaller models would miss. This scale allows RiNALMo to make more accurate predictions about RNA structure and behaviour, helping scientists gain insights that previously required years of laboratory work.

When tested, RiNALMo demonstrated state-of-the-art performance across multiple benchmark tasks, surpassing existing AI models in predicting RNA structure and function.

“Think of RiNALMo as teaching a model to read a foreign language through countless sentences, and then asking it to write poetry, detect grammar, and tell stories in that language. That’s what RiNALMo does, with RNA,” said Prof Mile Sikic, Assistant Director, AI & Compute at A*STAR GIS and Professor at the University of Zagreb’s Faculty of Electrical Engineering and Computing.

“At A*STAR GIS, we are focused on pushing the frontiers of genomics and AI to develop a new generation of RNA-targeted or RNA-based drugs,” said Dr Wan Yue, Executive Director at A*STAR GIS and co-researcher of the RiNALMo project. “RiNALMo has demonstrated that it is not only possible to learn the language of RNA, but also to harness that understanding in powerful new ways, from predicting function to designing RNA molecules with therapeutic potential. This opens exciting opportunities for the development of next-generation RNA-based medicines.”

“Since AlphaFold, an AI system, showed how AI can advance 3D protein structure prediction, the race has been on to develop newer and better models in uncharted areas of biomedical research. By incorporating domain expertise from RNA structural modelling into new AI architecture designs, the team has been able to make a big leap forward in making RNA druggable and improving RNA drugs.", Dr Sebastian Maurer-Stroh, Executive Director of A*STAR BII.

This Singapore-led research published in Nature Communications on 1 July, exemplifies how local innovation can lead to improved AI models in biology, pushing the boundaries of what is possible in RNA science and its real-world applications.

###