Fig. 1

Overview of CellWalker. a Cells (circles) are connected based on similarity of their scATAC-seq profiles (purple edges). The network is extended using external data labels (e.g., expression levels of marker genes for a cell type, rectangles). These labels are connected to cells based on how well they correspond to each cell (e.g., fraction of cell’s peaks in marker gene promoters, red edges). Information is propagated across the combined network using global diffusion. b Diffusion results in an influence matrix that describes the information passed from each label and cell to every other label and cell. Label-to-label influence (purple) encodes label similarity, label-to-cell influence (teal) encodes which labels best describe each cell, cell-label influence (blue) can be used to map information encoded in accessibility back to labels, and cell-to-cell influence represents a label-influenced clustering of cells (light blue)