In the vast kingdom of plants, from mosses clinging to rocks to towering trees reaching the sky, lies the code of life, the nucleotides. Among them, RNA (ribonucleic acid) plays a crucial role in regulating plant growth, development, and adaptation to the environment, thereby significantly contributing to plant survival and biodiversity. In recent years, large language models have made remarkable advances in understanding human language. This progress in AI has inspired plant scientists to ask: if artificial intelligence can grasp the complexity of human language, could it also help decode the “language of life” in plants?
Recently, HFSP Fellowship Awardee Haopeng Yu and colleagues published a study in Nature Machine Intelligence titled “An Interpretable RNA Foundation Model for Exploration of Functional RNA Motifs in Plants”. The team developed PlantRNA-FM, an interpretable foundation model capable of learning the language of life in the plant kingdom and decoding functional RNA regulatory elements using AI. For the first time, PlantRNA-FM integrates both RNA sequence and RNA structure information from 1,124 plant species, covering a wide diversity of plants ranging from mosses to flowering species. Compared to existing DNA/RNA AI models, PlantRNA-FM demonstrated superior predictive performance on plant-specific tasks. For instance, in gene region annotation, the model achieved an F1 score of 0.974, far outperforming the previous best model with a score of 0.639.
Beyond prediction, PlantRNA-FM offers intuitive interpretability and successfully identifies critical RNA structural features that influence plant gene regulation. Prior to the advent of AI, researchers relied on bioinformatic tools to correlate RNA sequence features (such as high GC content or T-rich regions) with their biological functions. However, due to the astronomical number of possible base combinations, traditional tools struggled to resolve more complex questions, such as which specific GC arrangements or T patterns truly impact function. Using PlantRNA-FM’s interpretability framework, scientists identified 112 translation-related RNA structural motifs, including 63 that repress translation and 49 that enhance it. Experiments confirmed that altering these structures can significantly affect translation efficiency by up to 5.3-fold.
PlantRNA-FM helps us understand how RNA, through both its sequence and structure, governs essential biological processes in plants. This paves the way for optimising gene expression and designing next-generation crops to address climate change and food security challenges. By decoding the language of plant life with AI, this study also highlights the transformative potential of interdisciplinary research in advancing the life sciences.