In a new article from Nature, researchers from UHN’s Toronto General Hospital Research Institute (TGHRI) shared their vision for how artificial intelligence (AI) could transform the way scientists explore the inner workings of cells.
As techniques like genomics and proteomics (large-scale studies of genes and proteins to understand biological systems) generate high volumes of biological data, researchers are looking for tools to help make sense of it all. Inspired by large language models like ChatGPT, scientists are now aiming to create ‘multimodal foundation models or MFMs'—AI models that can understand and process different types of information, like text, images, and numbers, all at the same time—for biological data. These models could be trained on many types of data, including DNA, RNA, protein, and the spatial organization of cells.
Dr. Bo Wang, Chief AI Scientist at UHN and Senior Scientist at TGHRI, and his colleagues believe that by integrating large biological datasets, MFMs could give researchers the ability to predict cell types, identify gene functions, and understand gene regulation across different tissues and disease states. By breaking down DNA, RNA, and protein data into small chunks, similar to how large language models process words and phrases, AI systems can learn to understand both the fine details, such as individual genes, and the bigger picture, such as how whole systems interact.
However, creating these powerful models requires significant amounts of high-quality data, as well as advanced computing power. Although the technology holds tremendous promise, challenges remain, including risks like “hallucination”, where AI produces incorrect but plausible results.
Despite these hurdles, Dr. Wang and his colleagues aim to build flexible models that can perform multiple tasks, ranging from simulating gene activity to predicting cell behaviour under specific conditions, such as genetic changes or drug treatments. The researchers believe that this approach could herald a new era in biomedical research, where AI plays a crucial role in decoding the complexities of life.
Dr. Haotian Cui, former doctoral student in Dr. Bo Wang’s lab, is the first author of the study.
Dr. Bo Wang, a Senior Scientist at Toronto General Hospital Research Institute and Assistant Professor in the Departments of Computer Science and Laboratory Medicine & Pathobiology at the University of Toronto, is the co-senior author of the study.
Dr. Fabian J. Theis, the Head of the Computational Health Center, Director of the Institute for Computational Biology at Helmholtz Munich, is the co-senior author of the study.
This work was supported by UHN Foundation.
Dr. Fabian J Theis consults for Immunai, CytoReason, Cellarity, BioTuring and Genbio AI, and has an ownership interest in Dermagnostix GmbH and Cellarity. Dr. Bo Wang serves as a scientific advisor to Shift Bioscience, Deep Genomics and Vevo Therapeutics, and acts as a consultant for Arsenal Bioscience.
See the manuscript for additional competing interests.
Cui H, Tejada-Lapuerta A, Brbić M, Saez-Rodriguez J, Cristea S, Goodarzi H, Lotfollahi M, Theis FJ, Wang B. Towards multimodal foundation models in molecular cell biology. Nature. 2025 Apr;640(8059):623-633. doi: 10.1038/s41586-025-08710-y. Epub 2025 Apr 16. PMID: 40240854.