UQ researchers have developed FeatureMAP, a machine-learning framework that helps scientists analyse how cells transition between states and identify genes associated with those changes, providing new insights into biological development, immune responses and disease.
What's the problem?
Advances in single-cell sequencing allow researchers to study the behaviour of individual cells in unprecedented detail. However, these datasets are highly complex, capturing the activity of thousands of genes across large numbers of cells.
Existing data visualisation and manifold-learning methods such as UMAP and t-SNE can reveal clusters of similar cells, but often do not preserve the underlying gene-level information. This can make it difficult to determine which genes characterise cellular processes, how cells transition between states, and which regulatory mechanisms may underpin development and disease.
What's UQ's solution?
Led by Dr Yang Yang from UQ’s Frazer Institute, researchers developed FeatureMAP (Feature-Preserving Manifold Approximation and Projection), a machine learning framework designed to make single-cell data analysis more interpretable.
FeatureMAP preserves both the relationships between cells and gene-level variation within a low-dimensional representation of biological data. This allows researchers to visualise not only how cells are related, but also how individual genes are associated with cellular states and transitions.
The method introduces new analytical concepts, including gene contribution, gene variation trajectories, and core and transition cell states, helping researchers identify important biological pathways and regulatory genes that may drive cellular transitions.
What's the impact?
FeatureMAP provides researchers with a more interpretable way to analyse single-cell data, helping uncover biological mechanisms that may be overlooked by existing approaches.
The method has been demonstrated using synthetic datasets and real-world single-cell RNA sequencing data, including studies of pancreatic development and CD8+ T-cell exhaustion, helping researchers analyse cellular trajectories, identify transition states and highlight potentially important regulatory genes.
By helping researchers identify key genes, cellular trajectories and transition states, FeatureMAP provides a powerful new tool for studying biological development, immune responses and the molecular mechanisms that underpin health and disease.
Did you know?
FeatureMAP can analyse gene contributions, cellular trajectories and transition states within the same framework, helping researchers identify regulatory genes associated with important biological processes.
Led by
Key collaborators
- Mr Jialei Gong
- Mr Hongjian Sun
- Mr Amos Choo
- Professor Jessica Mar
- Associate Professor Kelvin Tuong
- Professor Di Yu
| AI Research Strength | Data-Centric AI |
|---|---|
| Industry Portfolio | Health and Biomedical Sciences |
| Key Publication | Yang, Y., Gong, J., Sun, H. et al. Feature-preserving manifold approximation and projection to analyze single-cell data. Nat Comput Sci 6, 478–496 (2026). https://doi.org/10.1038/s43588-026-00970-6 |