Chatomics Field GuideWhat They Don't Teach You

Assay

Single-Cell ATAC-seq: sanity checks and pitfalls

Each cell yields a few thousand Tn5 fragments, so the peak-by-cell matrix is far sparser than a gene-by-cell matrix and is usually modeled with TF-IDF plus singular value decomposition (LSI) rather than log-normalization plus PCA. The first LSI component tracks sequencing depth and must be dropped; peaks called on the pooled data miss rare cell types unless you call per cluster. Cell calling, doublet detection and gene activity scores all have assay-specific pitfalls that scRNA-seq experience does not cover.

Who this is for: Single-cell analysts extending an scRNA-seq project with the 10x ATAC or Multiome kit. They expect Seurat-style workflows and hit a sparser, binary-ish, peak-by-cell matrix where PCA, normalization and marker calling all need different tools and intuitions.