Fig. 4: Velocity uncertainty and permutation score analysis. | Nature Methods

Fig. 4: Velocity uncertainty and permutation score analysis.

From: Deep generative modeling of transcriptional dynamics for RNA velocity analysis in single cells

Fig. 4: Velocity uncertainty and permutation score analysis.

a, From left to right: velocity stream, intrinsic and extrinsic uncertainty of the pancreas dataset estimated by veloVI. The left UMAP embedding is colored by cell type according to original cluster annotations11,35. Uncertainty is defined as the variance of the cosine similarity between samples of the velocity vector (intrinsic) and future cell states (extrinsic) and their respective mean (Methods). b, The corresponding cumulative distribution functions (CDFs) of the gene velocity coherence score is shown for alpha and Ngn3-high EP cells. The velocity coherence, defined for one cell and gene as the product of the velocity and the expected displacement of that cell/gene, is averaged within cell types. w.r.t., with regard to. c, The effect on the error between inferred dynamics and data when permuting unspliced and spliced abundance in the case of Top2a (top) and Sst (bottom). Coloring of cell types is according to a. d, Permutation score densities of datasets of the pancreas35, hippocampus8, forebrain8, spermatogenesis36, retina43, brain39, prefrontal cortex38, PBMCs and simulated data15 (top). Kurtosis (left) and skew (right) for each dataset (bottom).

Back to article page