Call peaks and find enriched motifs from ChIP-seq or ATAC-seq

Hand Claude Code aligned ChIP-seq or ATAC-seq BAMs; get back a called-peak BED, peaks annotated with their nearest genes, and a ranked list of enriched transcription-factor motifs.

   
Problem class Data analysis
Subject areas Molecular and Cellular Biology, Immunology and Microbiology
Evidence level Proposed
Complexity Multi-tool harness
Availability Fully open
Compute Laptop (no GPU)

Problem

After aligning a ChIP-seq, CUT&RUN, or ATAC-seq experiment and confirming the signal looks right, the substantive analysis is two coupled steps: call the enriched regions (where is the protein bound / where is the chromatin open?), then ask which transcription factors explain those regions (de novo and known-motif enrichment, plus nearest-gene context). MACS3 and HOMER are the field-standard pair, but they are a chain: MACS3 callpeak with the right narrow/broad mode and genome size, then HOMER findMotifsGenome.pl against a matched background and annotatePeaks.pl for gene context — and the peak file from the first has to be handed cleanly to the second. Solved looks like: BAMs in; a narrowPeak/broadPeak BED, a peak-to-gene annotation table, and a known/de-novo motif report out, with the peak mode and background choices explicit.

  1. Install the MACS3 and HOMER skills. Both ship in the SciAgent-Skills collection — clone once and load as a plugin:

    git clone https://github.com/jaechang-hits/SciAgent-Skills

    Then inside Claude Code run /plugin install sciagent-skills and confirm macs3-peak-calling and homer-motif-analysis appear under /plugin. Each skill installs its own dependencies (MACS3; HOMER + a genome FASTA) on first use.

  2. Prepare inputs. Provide coordinate-sorted, indexed BAMs (*.bam + *.bam.bai) for the IP/treatment and, ideally, a matched input/IgG control. Decide the peak mode up front: narrow for TF ChIP-seq and ATAC-seq, broad for H3K27me3/H3K9me3 and other broad histone marks. State the genome build so the effective genome size is correct.

  3. Call peaks with MACS3. A minimal prompt:

    Use the macs3 skill. Run callpeak on ip.bam with control input.bam,
genome size hs (GRCh38), narrow mode, q-value 0.05. Write the
narrowPeak BED to peaks/ and report the peak count and FRiP if
available.

    For a broad mark, say “broad mode, broad-cutoff 0.1” instead.

  4. Annotate and find motifs with HOMER. Continue:

    Use the homer skill. Run annotatePeaks.pl on peaks/ip_peaks.narrowPeak
against hg38 to assign nearest gene and TSS distance, then
findMotifsGenome.pl on the same peaks vs hg38 (size 200) for known
and de novo motif enrichment. Save the motif report to motifs/ and
summarize the top enriched TFs.

  5. Sanity-check the motif against the assay. For a TF ChIP-seq, the factor’s own motif should top the enrichment list — if it doesn’t, suspect the antibody, the peak set, or the background before interpreting downstream TFs.

Why this assembly

Rung 3 of the simplicity ladder — a two-component toolbelt. Rung 2 (a single skill) doesn’t solve it: peak calling and motif analysis are genuinely different tools (MACS3’s Poisson model vs HOMER’s motif enrichment), and the value is in the handoff — MACS3’s peak BED becomes HOMER’s input. Both skills come from the same collection and are designed to chain (HOMER’s catalog page explicitly says “use after MACS3”), so the harness is two skills, not three. No autonomous system is warranted: the workflow is a fixed, well-understood two-step pipeline, not a generated multi-stage analysis. This recipe is the binding-site/motif companion to the signal-profiling recipe, which deliberately stops before peak calling.

Availability

Fully open. Both skills are OSS in jaechang-hits/SciAgent-Skills (CC BY 4.0); MACS3 is BSD-3-Clause and HOMER is GPL-3.0. HOMER requires downloading a genome FASTA/annotation package on first use (free). No subscription or institutional access.

Compute requirements

Laptop-sufficient; no GPU. MACS3 callpeak on a typical mammalian ChIP-seq sample (20–40 M reads) runs in a few minutes on 4–8 GB RAM. HOMER findMotifsGenome.pl is the heavier step — de novo motif discovery over a few thousand peaks takes 10–30 minutes single-threaded and benefits from -p for parallelism; the one-time HOMER genome package download is a few GB of disk. annotatePeaks.pl is fast. Move to a workstation only for very large peak sets or many samples.

Evidence

Proposed. No documented attempt at an LLM-driven (Claude + MACS3 + HOMER) peak-calling-plus-motif workflow is known. The grounding is component-level and canonical: MACS is the standard model-based ChIP-seq peak caller (Zhang et al., “Model-based Analysis of ChIP-Seq (MACS),” Genome Biology 9:R137 (2008)), now maintained as MACS3, and HOMER is the standard motif-discovery and peak-annotation suite (Heinz et al., Molecular Cell 38:576 (2010)). The MACS3 → HOMER chain is the textbook ChIP-seq/ATAC-seq downstream pipeline, and both skills were evaluated as part of the BixBench-benchmarked SciAgent-Skills collection (see the MACS3 and HOMER catalog entries). The closest documented LLM analog is the cataloged Biomni agent composing genomics CLIs autonomously; this recipe pulls that to the lowest rung that solves the stated two-step problem.

Alternatives considered

  • The signal-profiling recipe (deepTools, rung 2). Reach for it when you already have a peak set (or TSS list) and want normalized coverage tracks and feature-centered heatmaps — visualization, not discovery. This recipe answers the upstream question of where the peaks are and what binds there.
  • Plain Claude Code shelling out to MACS3/HOMER. Fine for a one-off audit of every flag, but you re-derive the narrow/broad mode, background, and gene-annotation conventions each time.
  • A Nextflow/Snakemake pipeline (nf-core/chipseq, nf-core/atacseq). The right choice for production cohorts run repeatedly on a cluster — see the Nextflow development skill. This recipe is for the interactive “I have BAMs, find the peaks and the motifs” case.

See also

Sources

Tried this recipe?

Share feedback — what worked, what didn’t, what you’d change. The form opens with this recipe pre-selected and a link back to this page.