ATAC-seq peaks in the PBMC dataset were identified using MACS2 (ref. 46) with the following arguments: -g 2.7e9 -f BED – nomodel –extsize 200 –shift −100 –max-gap 50. We used the fragment file as input to the peak calling algorithm, as this contained the deduplicated Tn5 insertion sites for each cell. Peak calling was performed for each cell type using the CallPeaks function in Signac, with group.by = ‘celltype’ to call peaks on each predicted cell type separately and combine the resulting peak calls across all cell types. We removed any peaks overlapping annotated genomic blacklist regions for the hg38 genome61. We quantified counts for the resulting peak set for each cell using the FeatureMatrix function in Signac.
