Where this lecture fits

• Another bonus upstream module — it runs before the core scATAC analysis
• The RNA companion (Lecture 11) covered cellranger count; this is its chromatin twin
• Same SLURM patterns, same FASTQ rules, same reference model — different outputs
• The hand-off target is Module 15 (Signac), not Module 01

Companion materials

• Reading: Chapter 12 — Raw scATAC-seq Processing
• Tutorial: Tutorial 12 — Raw scATAC-seq with cellranger-atac
• RNA companion: Lecture 11 — Cell Ranger: Raw Reads to Counts
• Downstream: Lecture 15 — scATAC-seq with Signac

Goals of this lecture

• See how scATAC differs from scRNA at the data level (peaks vs genes, fragments vs UMIs)
• Run cellranger-atac count and read its parameters
• Pick the right reference (arc vs gex vs atac)
• Read ATAC-specific QC: TSS enrichment, FRiP, nucleosome signal
• Know which file feeds Module 15: fragments.tsv.gz
• Note when to switch to multiome (cellranger-arc)

Different measurement, parallel pipeline

• scRNA measures transcription — polyA mRNA captured, counted as UMIs per gene
• scATAC measures accessibility — Tn5-cut DNA fragments, counted per genomic peak
• The plumbing (SLURM, FASTQ naming, reference model) is nearly identical
• The outputs diverge because the underlying biology is different

RNA vs ATAC, side by side

The new file: fragments.tsv.gz

• A sorted, gzipped TSV — one row per Tn5-cut fragment
• Columns: chrom start end barcode duplicate-count
• This is the file Signac and ArchR consume directly
• It travels with its index: fragments.tsv.gz.tbi — never lose it in transit

The core invocation

cellranger-atac count \
  --id=atac_pbmc_1k_v1 \
  --reference=/projects/<PIRG>/<user>/refs/refdata-cellranger-arc-GRCh38-2020-A-2.0.0 \
  --fastqs=/projects/<PIRG>/<user>/test_data/atac_pbmc_1k_v1_fastqs \
  --sample=atac_pbmc_1k_v1 \
  --localcores=16 \
  --localmem=120

• Note --reference, not --transcriptome — the flag name differs from RNA
• No --create-bam flag here; a position-sorted BAM is written by default
• Same --localcores / --localmem discipline as RNA (budget below SLURM --mem)

Why it runs 2–3× longer than RNA

• Aligns fragments to the whole genome, not a small transcriptome
• Deduplicates at single-fragment resolution
• Calls peaks de novo with a MACS-style algorithm — RAM-hungry
• Counts fragments per peak per barcode
• Plan your SLURM --time accordingly (6–10 h is typical)

arc vs gex vs atac — choose carefully

• refdata-gex-GRCh38-... — RNA only. Do not pass this to cellranger-atac
• refdata-cellranger-arc-GRCh38-2020-A-2.0.0 — the current ATAC / multiome reference
• refdata-cellranger-atac-GRCh37-1.2.0 — older hg19, for matching EnsDb.Hsapiens.v75 in Module 15
• They share fasta/ and genes/ folders but have different indexes and annotations

Download the ATAC reference

cd /projects/<PIRG>/<user>/refs

# Current human ATAC / arc reference
wget https://cf.10xgenomics.com/supp/cell-atac/refdata-cellranger-arc-GRCh38-2020-A-2.0.0.tar.gz
tar -xzf refdata-cellranger-arc-GRCh38-2020-A-2.0.0.tar.gz

• The ATAC reference is separate because it indexes the whole genome, not just transcribed regions
• A wrong-reference run usually fails to start — but the error can be cryptic

The outputs are different

outs/
  ├── web_summary.html
  ├── summary.csv
  ├── singlecell.csv             per-barcode QC
  ├── filtered_peak_bc_matrix.h5  peaks × barcodes
  ├── fragments.tsv.gz (+ .tbi)   the key file for Signac
  ├── peaks.bed                   peak coordinates
  └── possorted_bam.bam (+ .bai)

Metrics to scrutinize

What these metrics mean

• TSS enrichment — fragments concentrate at transcription start sites in healthy cells; low = high background
• FRiP (fraction of reads in peaks) — how much signal lands in real open regions vs noise
• Nucleosome signal — the fragment-size profile shows nucleosome-free (<100 bp), mono- (~200), di- (~400) peaks
• A flat fragment-size distribution signals poor digestion or contamination

Four files Module 15 expects

SRC=<user>@login.talapas.uoregon.edu:/projects/.../outs

scp $SRC/filtered_peak_bc_matrix.h5  data/
scp $SRC/singlecell.csv              data/
scp $SRC/fragments.tsv.gz            data/
scp $SRC/fragments.tsv.gz.tbi        data/

• cellranger-atac produces exactly what Module 15 loads — no reformatting needed
• Module 15 uses Read10X_h5() + CreateChromatinAssay()
• The .tbi index must travel with fragments.tsv.gz or Signac can’t read it

Scaling and gotchas

• Same SLURM array pattern as RNA — one sample per row of samples.txt
• Wrong reference (gex instead of arc) is the #1 failure — cryptic error
• Lost .tbi breaks Signac silently downstream
• Mixed chemistries (v1 vs v1.1 Next GEM) in one folder — different barcode conventions
• OOM in peak calling — the de-novo step needs ~80 GB for 10k cells

Reference version and annotation matching

• The tutorial (Module 15) uses the hg19 PBMC 10k v1 dataset and therefore needs EnsDb.Hsapiens.v75
• If you ran cellranger-atac against refdata-cellranger-arc-GRCh38-2020-A-2.0.0 (hg38), use EnsDb.Hsapiens.v86 in Signac — chromosome coordinate offsets between hg19 and hg38 are large; mixing them produces bizarre errors
• Always confirm: what genome did the fragment file come from? Pick the matching EnsDb version accordingly
• To convert between builds use rtracklayer::liftOver() with a UCSC chain file (hg19ToHg38 or vice versa) — covered in the Signac tutorial

When RNA and ATAC come from the same cells

• A 10x Multiome library captures RNA and ATAC from each cell
• Use cellranger-arc count (not -atac) — it produces both a counts matrix and a fragments file
• Outputs are pre-paired by barcode — RNA and ATAC for the same cell are linked
• Same SLURM pattern; bigger reference; longer runtime
• This is what enables joint RNA+ATAC analysis downstream

What to take away

• scATAC measures accessibility (fragments/peaks), not transcription (UMIs/genes)
• Use cellranger-atac count with --reference, the arc (not gex) reference
• It runs longer because of whole-genome alignment + de-novo peak calling
• Read TSS enrichment, FRiP, and nucleosome signal — ATAC’s own QC vocabulary
• fragments.tsv.gz (+ .tbi) is the hand-off to Module 15 (Signac)
• Joint RNA+ATAC → switch to cellranger-arc

Next steps

• Read: Chapter 12 — Raw scATAC-seq Processing
• Do: Tutorial 12 — cellranger-atac on Talapas
• Then: Lecture 15 — scATAC-seq with Signac
• RNA companion: Lecture 11 — Cell Ranger: Raw Reads to Counts