Fig. 2
From: DeepVelo: deep learning extends RNA velocity to multi-lineage systems with cell-specific kinetics
Fine-grained temporal patterns in neurogenesis predicted by DeepVelo. a Comparison of DeepVelo with the dynamical model from scVelo [2]. The direction and magnitude of velocities are projected as arrows onto the Uniform Manifold Approximation and Projection (UMAP) plot of gene expression values across cells. DeepVelo provides more consistent velocity estimates with respect to the developmental process from immature granule cells to mature granule cells. b The boxplot and histogram of the overall consistency scores for scVelo and DeepVelo, which indicate the consistency of velocity estimates in a local neighborhood of the data. c The box plot and histogram of the cluster/cell-type-specific consistency scores, which utilize the neighborhood consistency metric on a per cluster/cell-type basis. d, e The spliced/unspliced phase portrait for Tmsb10 and Ppp3ca, respectively. Cell-types are shown in the same color as in b. f, g Velocity and gene expression values projected onto UMAP plots for Tmsb10 and Ppp3ca, respectively. Velocity and gene expression values show consistent patterns across cell-types: high velocity values (green in velocity plot) are correctly shown in the subset of cells developing to high gene expression values (purple in expression plot)