Abstract
High-dimensional single cell profiling coupled with computational modeling holds the potential to elucidate developmental sequences and define genetic programs directing cell lineages. However, existing algorithms have limited ability to elucidate branching developmental paths or to identify multiple branch points in an unsupervised manner. Here we introduce the concept of “trajectory space”, in which cells are defined not by their phenotype but by their distance along nearest neighbor trajectories to every other cell in a population. We implement a tSpace algorithm, and show that multidimensional profiling of cells in trajectory space allows unsupervised reconstruction of developmental pathways, and in combination with existing biological knowledge can be used to infer the identity of progenitor populations and of the most differentiated subsets within samples. Applied to high dimensional flow and mass cytometry data, the method faithfully reconstructs known branching pathways of thymic T cell development, and reveals patterns of tonsillar B cell development and of B cell migration. Applied to single cell transcriptomic data, the method unfolds the complex developmental sequences and genetic programs leading from intestinal stem cells to specialized epithelial phenotypes. Profiling of complex populations in high-dimensional trajectory space should prove useful for hypothesis generation in developing cell systems.
Footnotes
↵‡ shared authorship