We downloaded processed scATAC-seq counts from Zenodo (https://zenodo.org/record/3977808) and the fragment file from NCBI Gene Expression Omnibus (GSE148509) and computed the nucleosome signal and TSS enrichment score per cell as described above and retained cells with >1,000 counts and <50,000 counts, <5% of counts in genomic blacklist regions, TSS enrichment score >3 and a nucleosome signal score <4 and a mitochondrial genome sequencing depth of ≥10. We performed dimension reduction using LSI and UMAP as described above and identified clusters using the Smart Local Moving algorithm using the FindClusters function in Seurat with resolution = 0.5 and algorithm = 3 (ref. 62).
