peak overlapped multiple Cellranger peaks (Supplementary Fig. 1c). This revealed a bias in Cellranger for aberrant merging of multiple distinct peaks into a single region and highlights the importance of accurate cell-type-specific peak calling methods in the analysis of single-cell chromatin datasets. In the absence of multimodal data, an independent clustering of the cells can be performed using the chromatin data and peaks identified per cluster in place of cell-type-specific peak calling. To assess the similarity between cluster-specific and cell-type-specific peak calls, we clustered the cells using the DNA accessibility assay (Supplementary Fig. 2) and called peaks per cluster using MACS2. We found that 92.7% of cell-type-specific peaks overlapped cluster-specific peaks, while only 78.5% of cell-type-specific peaks overlapped a peak identified using the bulk-cell data. Finally, to evaluate how the size of a cell population influenced the ability to detect peaks in that population, we randomly sampled cells from the CD14+ monocyte population, with the total number of sampled cells ranging from 50 to 2,850 cells. For each downsampling, we called peaks using MACS2 and assessed the fraction of peaks identified using the 2,850-cell population that were able to be recovered using each downsampled population (Supplementary Fig. 1d.e). When sampling
