ChIP-seq measures where a protein sits on DNA; ATAC-seq measures where DNA is open. They share an analysis spine — align, filter, call peaks, annotate, quantify — but the biology behind a peak is entirely different, and conflating the two is the most common conceptual error in epigenomics. An ATAC peak means accessible chromatin, not bound factor; a ChIP peak means enriched occupancy, which may be indirect through a co-factor or epitope artifact.
The shared preprocessing spine
Both assays start from short reads and a reference genome. Trim adapters and low-quality bases, align with an aligner that handles indels, mark and remove duplicates, and keep only uniquely mapped, high-confidence reads. For human data, remove blacklist regions — hyper-mappable or artefactual loci that generate peaks in nearly every experiment.
- ChIP-seq needs an input or IgG control; the control is what makes peak calling statistically meaningful.
- ATAC-seq needs no control in the same sense, but its peaks are shaped by the Tn5 insertion bias, not by binding.
- Always inspect fragment-size distribution: ATAC should show the nucleosome-free sub- and mono-nucleosomal peaks.
ATAC-seq: assay design decides the result
The original Buenrostro protocol used a tagmentation reaction on fresh nuclei and produced workable but inconsistent signal-to-noise. Omni-ATAC improved it with a detergent-based lysis that strips the mitochondrial membrane and a Tween-containing transposition buffer, cutting the mitochondrial read fraction from routinely 60% or more down to single digits and sharpening the signal at regulatory elements. If your ATAC libraries are half mitochondrial reads, no computational trick will recover the lost coverage.
Quantify accessibility as read counts in a consensus peak set, then normalise for depth and for Tn5 bias. Counting in a fixed peak set makes samples comparable; discovering peaks per sample makes comparisons fragile unless you merge into a union set first and re-count. Accessibility is also a proxy for activity, not a direct measure of it — an open promoter may still be transcriptionally silent, and a closed one may be poised. Inferring transcription factor activity from accessibility alone, without expression or footprint evidence, is a common overreach.
Peak calling with MACS2
MACS2 is the canonical caller for ChIP-seq and remains usable for ATAC-seq. Its model-based approach slides a window across the genome, estimates the local background from a control, and computes an enrichment score; it then builds a dynamic Poisson model from the genome-wide distribution of background reads to assign p-values. The key idea is local lambda: background is estimated in a neighbourhood, not globally, so peaks in regions of high copy number or open chromatin are not systematically over-called.
| Assay | Caller guidance |
|---|---|
| ChIP-seq, sharp marks | MACS2 with matched input, q-value cutoff |
| ChIP-seq, broad marks (H3K27me3) | MACS2 --broad, or broad-domain callers |
| ATAC-seq | MACS2 or accessibility-specific callers, then union set |
Peak counts are not evidence of anything on their own. A million peaks typically means mis-normalised input or excessive sensitivity; a few hundred means the antibody failed. Run a spike-in control when comparing across conditions, because global changes in histone modification levels will otherwise be invisible to depth normalisation.
Motif discovery and factor occupancy
Motif analysis asks whether a discovered sequence pattern is enriched in your peaks, which links an unknown peak set to a candidate regulator. Two families of tools exist: de novo discovery, which infers motifs from the peaks themselves, and database matching against curated motif collections. Report the background you use — the choice of background sequence changes the enrichment ranking substantially, and repeating or GC-rich genomic background can manufacture spurious motif hits.
Footprinting goes further: within accessible regions, a bound factor protects its DNA from Tn5 or nuclease cleavage, leaving a dip in the cut-site pileup at the motif. Footprinting can infer factor occupancy from ATAC data alone, which is powerful but noisy — the footprint signal is weak relative to coverage requirements, and results should be cross-checked against ChIP or perturbation data rather than reported as definitive.
From peaks to function
Annotate peaks to genes, but be careful about the assignment rule. Nearest-gene assignment is fast and often wrong for distal regulatory elements, which may loop to a different gene entirely. Better approaches combine distance with contact evidence, expression correlation across samples, or chromatin-loop data. Assigning an enhancer to its nearest gene is a convenience, not a biological claim.
For multi-condition work, quantify peaks as a count matrix and test for differential accessibility with negative-binomial models analogous to RNA-seq — the statistics transfer directly. Then integrate: overlaps with differentially expressed genes, co-occurrence of motifs, and correlation with chromatin state are where epigenomic data becomes a mechanism rather than a catalogue.
Reproducibility and the small-effect trap
Epigenomic peaks are not supremely reproducible across labs and antibodies, and the same biological condition analysed against a different reference or blacklist version yields a partially different peak set. That is not necessarily error — it is resolution at the edge of signal. The honest response is to do everything within one pipeline, share processed matrices, and state the version of every reference. Two-fold accessibility changes near a low-coverage peak should not be reported as definitive biology when the underlying fragment counts are in the single digits.
References
- Zhang Y, Liu T, Meyer CA, et al. Model-based analysis of ChIP-Seq (MACS). Genome Biol. 2008;9(9):R137.
- Buenrostro JD, Giresi PG, Zaba LC, Chang HY, Greenleaf WJ. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat Methods. 2013;10(12):1213–1218.
- Corces MR, Trevino AE, Hamilton EG, et al. An improved ATAC-seq protocol reduces background and enables interrogation of frozen tissues. Nat Methods. 2017;14(10):959–962.
- Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30(15):2114–2120.
- Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods. 2012;9(4):357–359.
- Bailey TL, Boden M, Buske FA, et al. MEME Suite: tools for motif discovery and searching. Nucleic Acids Res. 2009;37(Web Server issue):W202–W208.
- Amemiya HM, Kundaje A, Boyle AP. The ENCODE Blacklist: identification of problematic regions of the genome. Sci Rep. 2019;9(1):9354.
- Landt SG, Marinov GK, Kundaje A, et al. ChIP-seq guidelines and practices of the ENCODE and modENCODE consortia. Genome Res. 2012;22(9):1813–1831.