MACS3 — ChIP-seq and ATAC-seq Peak Caller
SkillAI & modelsPoisson-model peak caller for ChIP-seq/ATAC-seq BAMs. MACS3 callpeak finds enriched regions (TF sites or histone marks) vs input/IgG; outputs BED narrowPeak/broadPeak for motif analysis, annotation, and differential binding. Use narrow peaks for TF ChIP-seq and ATAC-seq; broad for H3K27me3, H3K9me3, and other broad marks.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the MACS3 — ChIP-seq and ATAC-seq Peak Caller skill
What this skill tells your AI
The instructions your AI receives, as published by jaechang-hits/sciagent-skills in skills/genomics-bioinformatics/macs3-peak-calling/SKILL.md and read by ahel’s review.
Overview
MACS3 (Model-based Analysis of ChIP-seq) identifies regions of significant read enrichment (peaks) from ChIP-seq, ATAC-seq, CUT&RUN, and CUT&TAG experiments. It models the fragment length distribution from paired-end data or estimates it from mono-nucleosomal read shifting in single-end data, then applies a Poisson model to identify fold-enrichment over an input/IgG control. MACS3 produces BED-format narrowPeak (for transcription factors) or broadPeak (for histone marks) files with signal and q-value tracks for visualization in IGV or UCSC Genome Browser.
When to Use
- Calling transcription factor binding peaks from ChIP-seq experiments (use
--nomodel --extsize 200or let MACS3 estimate fragment length) - Identifying open chromatin regions from ATAC-seq experiments (use
--nomodel --shift -100 --extsize 200 -f BAMPE) - Calling broad histone modification peaks (H3K27me3, H3K9me3, H3K36me3) with
--broad - Generating peak signal tracks (bedGraph/bigWig) for genome browser visualization with
-B --SPMR - Performing differential binding analysis: MACS3 peaks as input to DiffBind or DESeq2
- Use HMMRATAC (part of MACS3) for nucleosome-resolution ATAC-seq peak calling
- Use SPP or HOMER as alternatives; MACS3 is the ENCODE-recommended standard
Prerequisites
- Python packages:
macs3(Python ≥ 3.8) - Input: Sorted BAM files (with index) from ChIP-seq or ATAC-seq alignment (e.g., using STAR or Bowtie2)
- Optional: Input/IgG control BAM for background normalization
Check before installing: The tool may already be available in the current environment (e.g., inside a
pixi/condaenv). Runcommand -v macs3first and skip the install commands below if it returns a path. When running inside a pixi project, invoke the tool viapixi run macs3rather than baremacs3.
# Install with pip or conda
pip install macs3
# or
conda install -c bioconda macs3
# Verify
macs3 --version
# macs3 3.0.2
Quick Start
# Call peaks for TF ChIP-seq (narrow peaks, with input control)
macs3 callpeak \
-t chip.bam \
-c input.bam \
-f BAM \
-g hs \
-n sample_tf \
--outdir peaks/ \
-q 0.05
# Output: peaks/sample_tf_peaks.narrowPeak
wc -l peaks/sample_tf_peaks.narrowPeak
Workflow
Step 1: Prepare Input BAM Files
MACS3 requires sorted, indexed BAM files from genome alignment.
# Sort and index ChIP and control BAMs (if not already done)
samtools sort -@ 8 chip_raw.bam -o chip.bam
samtools sort -@ 8 input_raw.bam -o input.bam
samtools index chip.bam
samtools index input.bam
# Check read counts
echo "ChIP reads: $(samtools view -c -F 4 chip.bam)"
echo "Input reads: $(samtools view -c -F 4 input.bam)"
Step 2: Call Narrow Peaks (TF ChIP-seq)
Use the default mode for transcription factor binding site identification.
# TF ChIP-seq with input control
macs3 callpeak \
-t chip.bam \
-c input.bam \
-f BAM \
-g hs \
-n tf_chip \
--outdir peaks/ \
-q 0.05 \
--keep-dup auto
echo "Peaks called: $(wc -l < peaks/tf_chip_peaks.narrowPeak)"
echo "Summit file: peaks/tf_chip_summits.bed"
# Without input control (less recommended)
macs3 callpeak \
-t chip.bam \
-f BAM \
-g hs \
-n tf_noinput \
--outdir peaks/ \
--nolambda
Step 3: Call Broad Peaks (Histone Marks)
Use --broad for spread histone modifications like H3K27me3 or H3K36me3.
# H3K27me3 broad histone mark
macs3 callpeak \
-t h3k27me3.bam \
-c input.bam \
-f BAM \
-g hs \
-n h3k27me3 \
--outdir peaks/ \
--broad \
--broad-cutoff 0.1 \
-q 0.05
echo "Broad peaks: $(wc -l < peaks/h3k27me3_peaks.broadPeak)"
# H3K4me3 (sharp mark — use narrow peaks)
macs3 callpeak \
-t h3k4me3.bam \
-c input.bam \
-f BAM \
-g hs \
-n h3k4me3 \
--outdir peaks/ \
-q 0.05
Step 4: Call ATAC-seq Peaks
ATAC-seq requires special handling for the Tn5 insertion site.
# ATAC-seq with paired-end BAM (recommended)
macs3 callpeak \
-t atac.bam \
-f BAMPE \
-g hs \
-n atac_sample \
--outdir peaks/ \
--nomodel \
--nolambda \
-q 0.05 \
--keep-dup all
echo "ATAC peaks: $(wc -l < peaks/atac_sample_peaks.narrowPeak)"
# Single-end ATAC-seq: shift reads to center on Tn5 cut site
macs3 callpeak \
-t atac_se.bam \
-f BAM \
-g hs \
-n atac_se \
--outdir peaks/ \
--nomodel \
--shift -100 \
--extsize 200 \
--keep-dup all
Step 5: Generate Signal Tracks for Visualization
Produce bedGraph and bigWig files for genome browser visualization.
# Generate bedGraph normalized to million reads (SPMR)
macs3 callpeak \
-t chip.bam \
-c input.bam \
-f BAM \
-g hs \
-n chip_track \
--outdir tracks/ \
-B \
--SPMR \
--keep-dup auto
# Convert bedGraph to bigWig for IGV/UCSC
# Requires bedGraphToBigWig and chrom.sizes
sort -k1,1 -k2,2n tracks/chip_track_treat_pileup.bdg > tracks/chip_sorted.bdg
bedGraphToBigWig tracks/chip_sorted.bdg genome/hg38.chrom.sizes tracks/chip.bw
echo "BigWig track: tracks/chip.bw"
Step 6: Annotate and Analyze Peaks
Parse narrowPeak output and annotate peaks to genomic features.
import pandas as pd
# Load narrowPeak file
# Columns: chrom, start, end, name, score, strand, signalValue, pValue, qValue, peak
cols = ["chrom", "start", "end", "name", "score", "strand",
"signalValue", "pValue", "qValue", "peak"]
peaks = pd.read_csv("peaks/tf_chip_peaks.narrowPeak", sep="\t",
header=None, names=cols)
print(f"Total peaks: {len(peaks)}")
print(f"Peaks on chr1: {(peaks['chrom'] == 'chr1').sum()}")
print(f"Median peak width: {(peaks['end'] - peaks['start']).median():.0f} bp")
print(f"Peaks with q-value < 0.01: {(peaks['qValue'] > 2).sum()}") # -log10(q) > 2
# Filter high-confidence peaks
high_conf = peaks[peaks["qValue"] > 2].copy() # q < 0.01
high_conf["width"] = high_conf["end"] - high_conf["start"]
print(f"\nHigh-confidence peaks: {len(high_conf)}")
high_conf.to_csv("high_confidence_peaks.bed", sep="\t", index=False, header=False,
columns=["chrom", "start", "end", "name", "score", "strand"])
Key Parameters
| Parameter | Default | Range/Options | Effect |
|---|---|---|---|
-t / --treatment | required | BAM/BED/SAM | ChIP or ATAC treatment file |
-c / --control | — | BAM/BED/SAM | Input/IgG control; omit --nolambda if absent |
-g / --gsize | required | hs, mm, ce, dm, or integer | Effective genome size; hs=2.7e9 (human), mm=1.87e9 (mouse) |
-q / --qvalue | 0.05 | 0–1 | FDR threshold for peak calling |
-p / --pvalue | — | 0–1 | P-value cutoff (use instead of q-value for strict control) |
--broad | off | flag | Call broad peaks for diffuse histone marks |
--broad-cutoff | 0.1 | 0–1 | Q-value cutoff for broad region merging |
--nomodel | off | flag | Skip fragment length modeling; required for ATAC-seq |
--extsize | 200 | 50–1000 | Fragment extension size when --nomodel is set |
--shift | 0 | -500–500 | Read shift in bp; use -100 with --extsize 200 for ATAC-seq |
--keep-dup | 1 | auto, all, integer | Duplicate handling; auto uses Poisson model, all keeps all (ATAC-seq) |
-B / --bdg | off | flag | Write bedGraph signal tracks |
--SPMR | off | flag | Normalize bedGraph to signal per million reads |
Common Recipes
Recipe 1: Batch Peak Calling for Multiple Samples
#!/bin/bash
# Call peaks for multiple ChIP-seq samples with the same input
INPUT="input.bam"
GENOME="hs"
OUTDIR="peaks"
mkdir -p "$OUTDIR"
SAMPLES=(H3K4me3 H3K27ac H3K27me3 CTCF)
MODES=(narrow narrow broad narrow)
for i in "${!SAMPLES[@]}"; do
sample="${SAMPLES[$i]}"
mode="${MODES[$i]}"
echo "Calling peaks: $sample ($mode)"
if [ "$mode" == "broad" ]; then
BROAD_FLAG="--broad --broad-cutoff 0.1"
else
BROAD_FLAG=""
fi
macs3 callpeak \
-t "${sample}.bam" \
-c "$INPUT" \
-f BAM \
-g "$GENOME" \
-n "$sample" \
--outdir "$OUTDIR" \
$BROAD_FLAG \
-q 0.05 \
--keep-dup auto
echo "$sample: $(wc -l < $OUTDIR/${sample}_peaks.*Peak) peaks"
done
Recipe 2: Reproducible Peaks with IDR (Irreproducible Discovery Rate)
# Call peaks on individual replicates (lenient thresholds for IDR)
for rep in rep1 rep2; do
macs3 callpeak \
-t "chip_${rep}.bam" \
-c input.bam \
-f BAM \
-g hs \
-n "tf_${rep}" \
--outdir peaks/ \
-p 0.1 \
--keep-dup auto
done
# Run IDR to find reproducible peaks
# pip install idr
idr --samples peaks/tf_rep1_peaks.narrowPeak peaks/tf_rep2_peaks.narrowPeak \
--input-file-type narrowPeak \
--output-file peaks/tf_idr_peaks.txt \
--idr-threshold 0.05 \
--plot
echo "IDR peaks: $(wc -l < peaks/tf_idr_peaks.txt)"
Expected Outputs
| Output | Format | Description |
|---|---|---|
*_peaks.narrowPeak | BED6+4 | Narrow peaks with signal, p-value, q-value, summit offset |
*_peaks.broadPeak | BED6+3 | Broad peaks (when --broad): chrom, start, end, signal, p-val, q-val |
*_summits.bed | BED3+2 | Peak summit positions (1 bp) with score; use for motif analysis |
*_treat_pileup.bdg | bedGraph | Treatment signal track (when -B) |
*_control_lambda.bdg | bedGraph | Control/local lambda track (when -B) |
*_model.r | R script | Fragment size model; run Rscript *_model.r to plot |
Troubleshooting
| Problem | Cause | Solution |
|---|---|---|
| Very few peaks called | Stringent q-value or low read depth | Relax to -p 1e-3; check sequencing depth (≥10M aligned reads recommended) |
| Too many peaks (>100k) | Threshold too loose or no input control | Add --control input.bam; use -q 0.01; filter on signalValue |
| Peak calling fails with "no reads" | BAM file is not sorted or indexed | Run samtools sort and samtools index before MACS3 |
| ATAC-seq peaks in mitochondria | High mtDNA content | Filter: `samtools view -h chip.bam |
| Fragment model fails | Too few reads or unusual read length | Add --nomodel --extsize 200 to skip modeling |
| bedGraph output very large | High coverage data without normalization | Add --SPMR to normalize to signal per million reads |
--broad misses narrow peaks | Signal is actually sharp | Check ChIP target: TFs and H3K4me3 need narrow mode |
gsize mismatch | Using wrong genome size for assembly | Use hs for hg19/hg38, mm for mm9/mm10; or provide exact integer |
References
- MACS3 GitHub: macs3-project/MACS — source code, documentation, and changelog
- Zhang Y et al. (2008) "Model-based Analysis of ChIP-Seq (MACS)" — Genome Biology 9:R137. DOI:10.1186/gb-2008-9-9-r137
- ENCODE ATAC-seq pipeline — ENCODE standardized ATAC-seq workflow using MACS3
- IDR framework — irreproducible discovery rate for reproducible peak calls
Signals
- GitHub stars
- 363
- Forks
- 36
- Last commit
- Aug 2026
Advanced
- Catalog kind
- skill
- Gateway key
macs3-peak-calling- Source
- github.com/jaechang-hits/sciagent-skills