Fig. 5: Continuum and embedding benchmarks against linear dimensionality reduction and manifold-learning methods.
From: iAODE for benchmarking and continuum modeling of single-cell chromatin accessibility
A–C For Small (n = 165), Medium (n = 68), and Large (n = 15) scATAC-seq datasets, iAODE is compared with ICA, FA, NMF, PCA, Diffusion Maps (DIFF), and Palantir. Each panel summarizes continuum/manifold-geometry metrics together with embedding-quality metrics computed from UMAP and t-SNE embeddings. Global Friedman or RM-ANOVA tests (all p < 0.001) indicate overall differences across methods. iAODE attains higher scores on Manifold dimensionality, Participation ratio, Noise resilience, Trajectory directionality, CAL, and COR than the linear and graph-based methods examined here, whereas Palantir and Diffusion Maps tend to yield higher ASW and QL, indicating relatively stronger emphasis on local compactness. DIFF Diffusion Maps, UMAP Uniform Manifold Approximation and Projection, t-SNE t-distributed stochastic neighbor embedding, ASW average silhouette width, DAV Davies--Bouldin index, CAL Calinski--Harabasz index, COR correlation-based coupling metric, DC distance correlation, QL local quality, QG global quality, OV overall embedding quality. Where shown, significance stars denote *p < 0.05, **p < 0.01, ***p < 0.001.
