AI Model Captures Hidden Signs of Disease at the Cellular Level

Researchers at McGill University have developed a machine learning tool that identifies disease-related patterns in RNA expression at a finer resolution than conventional methods. The tool, named DOLPHIN, focuses on exon-level data in individual cells and can detect biological signals that are often missed in gene-level analyses.

Published in Nature Communications, the study outlines how the model advances the detection of RNA splicing variations—changes in how RNA is assembled from building blocks called exons—revealing early signs of disease progression or severity.

Zooming in on RNA splicing

In cellular biology, many disease markers are expressed as subtle changes in RNA splicing. Traditional approaches aggregate RNA expression into overall gene counts, potentially obscuring differences that occur in specific segments. DOLPHIN instead analyses exon and junction reads, improving the resolution of single-cell transcriptomics and providing a clearer picture of molecular activity in health and disease.

In one analysis of pancreatic cancer samples, the researchers applied DOLPHIN to identify more than 800 exon-level markers that had not been detected by other tools. The model differentiated patients with high-risk forms of the disease from those with less aggressive tumors, based on these newly detected signals.

Toward more detailed digital models of cells

Beyond marker identification, the model can simulate how cells behave in response to treatment by constructing richer profiles of cellular states. These profiles allow researchers to generate “virtual cells”—in silico representations of real cells—that can be used to predict drug responses before laboratory or clinical testing.

The authors propose that this type of modelling could reduce time and cost in early-phase drug research by supporting hypothesis generation prior to experimentation.

The next phase of the research will involve scaling the tool to analyze millions of cells across diverse datasets. This expansion is intended to enhance the resolution and predictive capacity of virtual cell simulations, potentially improving disease modeling in future studies.

This work was supported by the Meakins-Christie Chair in Respiratory Research, the Canadian Institutes of Health Research, the Natural Sciences and Engineering Research Council of Canada, and the Fonds de recherche du Québec.

Reference: Song K, Zheng Y, Zhao B, et al. DOLPHIN advances single-cell transcriptomics beyond gene level by leveraging exon and junction reads. Nat Commun. 2025. doi: 10.1038/s41467-025-61580-w

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.