Reading Chromatin Accessibility: From ATAC-seq Peaks to Hypotheses

ATAC-seq is cheap, fast, and everywhere. The interpretive problem is that a peak file looks like a result: coordinates, scores, annotation. But three different questions get answered with the same output, and conflating them is where most accessibility papers go wrong.

The three questions are: where is chromatin open (a description), which open regions differ between conditions (a comparison), and which differences matter for transcription (a causal claim). This article is about moving from the first to the third without skipping the statistical and biological checks in between.

Note the scope: whether accessibility alone can establish enhancer function is a separate question, covered in our piece on ATAC-seq and functional enhancer identification. Here we assume the peaks are a starting point and focus on reading them correctly.

Four QC parameters determine whether the dataset is worth interpreting:

• • Fragment size distribution. Nucleosome-periodic fragments (sub-nucleosomal, mono-, di-nucleosomal) indicate successful transposition; a monotonic size distribution suggests over- or under-tagmentation.
• • TSS enrichment. Signal concentrated at transcription start sites is a quality signature; low enrichment usually means poor signal-to-noise rather than an unusual cell type.
• • Duplicate rate and library complexity. High duplication limits effective depth and inflates apparent differences.
• • Blacklist fraction. Signal in hyper-ChIPable regions indicates background; remove those regions before any comparison.

A sample that fails TSS enrichment is not rescued by deeper sequencing. This single check prevents a large share of unreliable downstream conclusions.

Peak callers differ in how they handle the nucleosome-sized gap in ATAC fragments, whether they call broad or narrow regions, and how they model local background. Three practical rules:

• • Use one caller consistently across all samples in a comparison. Merging peak sets from different callers creates artefacts that look like biology.
• • Call peaks separately, then build a union or consensus set for differential analysis, rather than using one sample's peaks as the reference for all.
• • Report parameters and depth. Peak number scales with depth; a comparison between libraries of very different depths needs subsampling or a normalization model.

For differential accessibility, the replicate unit is the donor or sample, never the cell. Treating cells as replicates inflates significance by orders of magnitude and is the most frequent statistical error in single-cell accessibility work.

Before assigning meaning, annotate honestly:

The distribution of these classes in your peak set is itself informative: a "regulatory landscape" dominated by promoter-proximal peaks is a transcriptome readout in disguise.

Footprints — dips in Tn5 insertion within accessible regions — can indicate factor occupancy. They are attractive because they suggest which motif is actually bound.

Caveats worth stating in the paper: footprint depth depends on sequencing depth and fragment-length filtering; nucleosomes and other complexes also produce footprints; and factor identity is inferred from motif, not demonstrated. Footprints are best used to rank candidate motifs for follow-up (for example, prioritization for base-editing validation), not to declare occupancy.

A defensible pipeline from peak to hypothesis:

1. Filter for interpretability. Remove blacklist, promoter-proximal, and repeat-derived regions if the question is enhancer-focused.
2. Require a second signal. H3K27ac, eRNA, or conservation — one supporting line turns a peak into a candidate active element.
3. Assign candidate targets explicitly. Contact data, activity-by-contact predictions, or colocalization. Label these as predictions.
4. Rank by expected effect. Strong differential accessibility plus active mark plus plausible target is the highest tier.
5. Select a testable subset. Typically 5–20 elements for a first perturbation round; the whole list is not the experiment.
6. State the falsifier. What result would make you abandon the hypothesis? If there is no such result, the hypothesis is not testable.

Step 6 is what distinguishes a functional study from a descriptive atlas.

EDITGENE builds the models that follow interpretation: CRISPRi-ready lines for element-level necessity testing, reporter lines for quantitative element activity, and matched cell backgrounds for comparing conditions without batch confounds.

Send us the peak set and the cell type. Contact EDITGENE to request a project consultation or quote.

Peak calls, differential results, and footprints depend on library quality, depth, caller, and normalization; conclusions should report these parameters.

Accessibility describes chromatin state, not regulatory function. Functional claims require perturbation with appropriate controls, and candidate target assignments are predictions until tested.

• • Buenrostro JD, et al. Transposition of native chromatin for fast and sensitive epigenomic profiling. Nature Methods. 2013;10:1213–1218.
• • Corces MR, et al. An improved ATAC-seq protocol reduces background and enables interrogation of frozen tissues. Nature Methods. 2017;14:959–962.
• • Vierstra J, et al. Global reference mapping of human transcription factor footprints. Nature. 2020;583:729–736.
• • Fullard JF, et al. An atlas of chromatin accessibility in the adult human brain. Cell. 2018;175:1–14.