Fig. 2: Leveraging pseudotemporal ordering for cellular fate mapping.
From: CellRank 2: unified fate mapping in multiview single-cell data
a, The PseudotimeKernel biases the edges of a phenotypic similarity-based nearest-neighbor graph toward increasing pseudotime, defining cell–cell transition probabilities. b,c, UMAP embedding of 24,440 peripheral blood mononuclear cells26, colored by cell type (cDC; G/M progenitor, granulocyte/myeloid progenitor; HSC; MK/E prog, megakaryocyte/erythrocyte progenitors; pDC). We illustrate the well-known differentiation hierarchy in black (b) as well as projected velocity fields based on the PseudotimeKernel (c, left) and RNA velocity (c, right). d, Correlating fate probabilities with gene expression recovers known lineage drivers for the pDC lineage27,28. We show lineage-specific trends as proposed in our earlier work14 by fitting generalized additive models to gene expression (y axis) in pseudotime (x axis); the contribution of each cell to each lineage is weighted according to CellRank 2-recovered fate probabilities. Colors correspond to lineages as shown in b.
