Skip to content

pipeline-chipseq.md

Latest commit

 

History

History
121 lines (89 loc) · 5.29 KB

pipeline-chipseq.md

File metadata and controls

121 lines (89 loc) · 5.29 KB

Pipeline: ChIP-seq -- ENCODE-Standard Processing from FASTQ to Peaks

Category: Pipeline Execution | Tools Used: encode_search_experiments, encode_list_files, encode_download_files, encode_log_derived_file

What This Skill Does

Runs a complete ENCODE-standard ChIP-seq pipeline via Nextflow DSL2: FASTQ quality control, BWA-MEM alignment, filtering and deduplication, MACS2 peak calling, IDR reproducibility analysis, and signal track generation. Outputs narrowPeak/broadPeak files, fold-change bigWig tracks, and a comprehensive QC report.

When to Use This

  • You have raw ChIP-seq FASTQs and need ENCODE-compliant peak calls and signal tracks.
  • You want to process your own data with the same pipeline used by the ENCODE DCC, then compare against ENCODE reference data.

Example Session

A scientist processes paired-end H3K27ac ChIP-seq from two biological replicates of human pancreatic islets with a matched input control.

Step 1: Download FASTQs from ENCODE

encode_list_files(
    experiment_accession="ENCSR831YAX",
    file_format="fastq", status="released"
)

encode_download_files(
    file_accessions=["ENCFF123ABC", "ENCFF456DEF", "ENCFF789GHI", "ENCFF012JKL"],
    download_dir="/data/chipseq/fastq", organize_by="flat"
)

Four FASTQ files downloaded: paired-end ChIP reads (R1/R2) and matched input control (R1/R2), all MD5-verified.

Step 2: Set Up Local Docker Execution

docker pull encodedcc/chip-seq-pipeline:v2.2.1

The nextflow.config ships with four profiles (local, slurm, gcp, aws). The local profile allocates 8 CPUs / 32 GB to BWA-MEM, 16 GB to deduplication, and 8 GB to MACS2. Machines under 32 GB RAM should use the SLURM profile.

Step 3: Run the Pipeline

nextflow run skills/pipeline-chipseq/scripts/main.nf \
  -profile local \
  --reads '/data/chipseq/fastq/chip_R{1,2}.fq.gz' \
  --control '/data/chipseq/fastq/input_R{1,2}.fq.gz' \
  --genome GRCh38 \
  --peak_type narrow \
  --outdir /data/chipseq/results

H3K27ac is a narrow active mark, so --peak_type narrow invokes macs2 callpeak --call-summits at q-value 0.05. For broad marks (H3K27me3, H3K36me3, H3K9me3), use --peak_type broad which adds --broad --broad-cutoff 0.1. The GRCh38 reference and ENCODE blacklist v2 auto-download on first run.

Step 4: Interpret QC Output

Open results/qc/multiqc/multiqc_report.html. Key metrics for a passing H3K27ac experiment:

Metric                      Rep1       Rep2       Threshold      Verdict
------------------------------------------------------------------------
Total reads                 48.2M      51.7M      >=20M          PASS
Mapping rate                96.3%      95.8%      >80%           PASS
Duplication rate            18.4%      21.1%      <30%           PASS
NRF                         0.87       0.84       >=0.8          PASS
NSC                         1.14       1.11       >1.05          PASS
RSC                         1.02       0.97       >0.8           PASS
FRiP                        8.2%       7.5%       >=1%           PASS
MACS2 peaks (q<0.05)       68,412     71,088      --             --
IDR peaks (0.05 threshold)  42,316     --         >20,000        PASS

FRiP at 7-8% indicates strong enrichment -- active histone marks typically range 5-15%, while most TF experiments fall at 1-5%. NSC/RSC above thresholds confirm real strand-shift signal. 42,316 IDR peaks shows excellent replicate reproducibility. Red flags: NRF below 0.8 (PCR bottleneck) or FRiP below 1% (weak enrichment or antibody failure).

Step 5: Key Output Files

Use IDR peaks as your primary peak set, and fold-change bigWig for browser visualization.

results/peaks/idr/idr_peaks.txt        # 42,316 reproducible peaks (use these)
results/peaks/narrow/                   # Per-replicate narrowPeak files
results/signal/fold_change/*.fc.bw      # Fold-change-over-control bigWig
results/signal/pvalue/*.pval.bw         # Signal p-value bigWig
results/qc/multiqc/multiqc_report.html  # Aggregated QC report

Step 6: Log Provenance

encode_log_derived_file(
    file_path="/data/chipseq/results/peaks/idr/idr_peaks.txt",
    source_accessions=["ENCSR831YAX"],
    description="IDR-filtered H3K27ac peaks from 2 replicates, ENCODE pipeline",
    file_type="idr_peaks",
    tool_used="ENCODE ChIP-seq pipeline v2.2.1 (BWA-MEM + MACS2 + IDR)",
    parameters="genome=GRCh38; peak_type=narrow; macs2 qvalue=0.05; idr threshold=0.05"
)

Cloud Cost Estimates

Platform Instance Cost/Sample Wall Time Notes
Local 8 cores, 32 GB $0 3-6 hours Docker required
GCP n1-standard-8 (preemptible) ~$2-5 2-4 hours Preemptible saves 60-80%
AWS m5.2xlarge (spot) ~$2-5 2-4 hours Spot instances recommended
SLURM 8 cores, 32 GB Varies 2-4 hours Singularity auto-mounted

A 2-replicate experiment (~50M reads each, ~40 GB total) stays under $10 on preemptible/spot.

Related Skills

  • pipeline-guide (parent) -- Pipeline selection and resource planning.
  • quality-assessment -- Deep-dive QC beyond the traffic-light summary.
  • histone-aggregation -- Merge peaks across experiments after peak calling.

Part of the ENCODE Toolkit -- 43 skills for genomics research