Fig. 3: False discoveries in single-cell DA analysis.

a Schematic overview of Experiment 3.1. Bone marrow mononuclear cells from healthy donors were profiled by Luecken et al.⁵⁹ in 13 independent replicates. For each cell type, half of these replicates were assigned to an artificial ‘control’ group, and the other half to an artificial ‘treatment’ group. DA analysis was then performed between cells from randomly assigned replicates. b Number of DA peaks detected between randomly assigned replicates at 5% FDR within each cell type in random comparisons of published scATAC-seq data (n = 21 comparisons). Inset text shows the median number of DA peaks per method. c As in b but showing the number of DA peaks detected at 5% FDR in DA analysis of downsampled bulk ATAC-seq libraries without biological differences between experimental conditions. d As in b but showing the number of DA peaks detected at 5% FDR in model-based simulations of scATAC-seq data without biological differences between experimental conditions. e As in b but showing the number of DA peaks detected at 5% FDR in model-based simulations of scATAC-seq data without biological differences between experimental conditions, with variation in sequencing depth between libraries. Source data are provided as a Source Data file (Source Data 3).