Back to publications

2026

Nat Comput Sci 2026 May;6(5):478-496. doi: 10.1038/s43588-026-00970-6. Epub 2026 Apr 17.

Feature-preserving manifold approximation and projection to analyze single-cell data

Yang Yang, Jialei Gong, Hongjian Sun, Amos Choo, Jessica Cara Mar, Yunbo Wei, Yu Zhang, Wenjie Zhang, Minglei Shu, Zewen Kelvin Tuong, Di Yu

Frazer Institute, Faculty of Health, Medicine and Behavioural Sciences, The University of Queensland, Brisbane, Queensland, Australia. Qilu University of Technology (Shandong Academy of Sciences), Jinan, China. Ian Frazer Centre for Children's Immunotherapy Research, Child Health Research Centre, Faculty of Health, Medicine and Behavioural Sciences, The University of Queensland, Brisbane, Queensland, Australia. Australian Institute for Bioengineering and Nanotechnology, The University of Queensland, Brisbane, Queensland, Australia. Laboratory of Immunology for Environment and Health, Shandong Analysis and Test Center, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China. School of Computer Science and Engineering, The University of New South Wales, Sydney, New South Wales, Australia.

Service type: Stock strains

Abstract

Visualizing single-cell data supports understanding cellular heterogeneity and dynamics. Uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbor embedding (t-SNE) reveal clustering structures but often fail to preserve underlying gene-level information. Here we introduce FeatureMAP (feature-preserving manifold approximation and projection), a framework that enhances manifold learning through pairwise tangent space embedding. By integrating UMAP with principal component analysis, FeatureMAP retains clustering structures and feature variation in a low-dimensional representation. It presents three key analytic concepts: gene contribution, gene variation trajectory, and core versus transition states. Gene contribution and gene variation trajectory are derived by estimating and projecting feature loadings or variation, whereas core and transition states are computationally defined using FeatureMAP's topological properties, including density, curvature and betweenness centrality. These concepts enable differential gene variation(DGV) analysis that highlights regulatory genes driving transitions between cell states. Demonstrated on synthetic and real single-cell RNA sequencing data from pancreatic development and T cell exhaustion, FeatureMAP improves analyses of trajectories and crucial regulatory genes.