Fig. 3

Whole-embryo RNA velocity and derived cell state transition map (A) Projection (UMAP) of zebrafish embryo transcriptomes (early development, 10 to 24 hpf time points), color coded by cell type. Arrow streams correspond to the average RNA velocities of single cells. On the right, stack bar plot of the cell-type proportion over time. (B) Workflow to reconstruct state transitions from RNA velocity graph. (1) Weighted directed graph represents the expected transitions between transcriptional states from single cells. (2) This graph defines a Markov process that can be represented as a matrix. (3) We can then estimate cluster transitions by simulating successive state transitions for groups of cells through the graph. (4) Finally, we summarize cluster-level transitions as a coarse-grained graph. (C) Examples of average transition probabilities for neural tube, endothelial cells, and mesodermal tissue. (D) A coarse-grain graph of cell state transitions shows transitions between all cell states. The width of the arrows is proportional to the transition rates. Colors and numbers correspond to the annotation in (A). Nodes grouped by a bounding box indicate clusters of cells that belong to the same broad cell types; colors correspond to those in (A). Very weak edges should be interpreted with caution, e.g., the connection between germline cells and endoderm might be spurious. See also Video S1C and Data S1.