/accessibility-aggregation

name: accessibility-aggregation
description: Build comprehensive chromatin accessibility maps by aggregating ATAC-seq and DNase-seq narrowPeak data across multiple ENCODE experiments, donors, and labs. Use when the user wants to answer "where is chromatin accessible in my tissue?" by combining peak calls into a union peak set. Handles cross-lab variation, ATAC vs DNase platform differences, and ENCODE blocklist filtering.

When to Use

• User wants to combine ATAC-seq or DNase-seq peaks across multiple experiments for a tissue
• User asks "where is chromatin accessible in my tissue?" or "build an open chromatin map"
• User needs to merge accessibility data from different labs, donors, or platforms (ATAC vs DNase)
• User wants a comprehensive set of open chromatin regions for regulatory element discovery
• Example queries: "aggregate ATAC-seq peaks for pancreas", "combine DNase-seq across donors", "find all accessible regions in liver"

Aggregate Chromatin Accessibility Peaks Across Studies

Build a comprehensive map of open chromatin for a tissue/cell type by merging ATAC-seq and/or DNase-seq narrowPeak files from multiple ENCODE experiments.

Scientific Rationale

**The question**: "Where is chromatin accessible in my tissue?"

Like histone marks, chromatin accessibility is a **detection question**. An open chromatin region detected in one donor but not another is still a real accessible site — individual variation, sequencing depth, and technical factors explain absence. We want the **union of all detections**.

ATAC-seq vs DNase-seq

Both measure open chromatin but with different biases:

| Property | ATAC-seq | DNase-seq | |----------|----------|-----------| | Method | Tn5 transposase insertion | DNase I hypersensitivity | | Input required | ~50K cells | ~1M cells | | Resolution | High | High | | GC bias | Moderate (Tn5 preference) | Low | | Mitochondrial reads | High (filter needed) | None | | ENCODE availability | Newer experiments | Extensive historical catalog | | Comparability | Generally comparable at open regions | |

Literature Support

• **Corces et al. 2017** (Nature Methods, 733 citations): Established that ATAC-seq and DNase-seq identify largely overlapping accessible regions, with ATAC capturing ~75% of DNase sites. Both are valid for union maps.
• **ENCODE Blacklist** (Amemiya et al. 2019, Scientific Reports, 1,372 citations): Comprehensive set of problematic genomic regions to filter. Essential for all functional genomics analyses. [DOI](https://doi.org/10.1038/s41598-019-45839-z)
• **F-Seq2** (Zhao & Boyle 2020, NAR Genomics): Improved peak caller for DNase-seq and ATAC-seq with proper test statistics for IDR compatibility.
• **ENCODE Phase 3** (Gorkin et al. 2020, Nature, 301 citations): Integrated accessibility data with histone marks across tissues for chromatin state annotation.

**Recommendation**: If combining ATAC-seq and DNase-seq peaks, treat them as equivalent signal sources for accessibility. The union is appropriate because both detect the same biological signal (open chromatin) through different enzymatic mechanisms.

Step 1: Find All Available Accessibility Data

# ATAC-seq
encode_search_experiments(
    assay_title="ATAC-seq",
    organ="pancreas",
    biosample_type="tissue",
    limit=100
)

# DNase-seq
encode_search_experiments(
    assay_title="DNase-seq",
    organ="pancreas",
    biosample_type="tissue",
    limit=100
)

Present a summary to the user:

• Total ATAC-seq experiments
• Total DNase-seq experiments
• Labs represented
• Whether to use one or both assay types

Combining ATAC + DNase?

Ask the user:

• **Same assay only** (purest comparison, no cross-platform effects)
• **Both assays combined** (maximum coverage, slight platform variation)

For a comprehensive accessibility catalog, combining both is scientifically justified.

Step 2: Quality-Gate Each Experiment

encode_get_experiment(accession="ENCSR...")

ATAC-seq Quality Checks

• Audit status: no ERROR flags
• Has IDR thresholded peaks
• Low mitochondrial read fraction (ENCODE pipeline removes these)
• Good TSS enrichment score
• Nucleosome-free fragment enrichment visible

DNase-seq Quality Checks

• Audit status: no ERROR flags
• Has Hotspot2 peaks or IDR thresholded peaks
• Adequate sequencing depth (20M+ mapped reads)
• Signal-to-noise ratio

Track all included experiments:

encode_track_experiment(accession="ENCSR...")

Step 3: Download Peak Files

For each experiment:

# ATAC-seq — IDR thresholded peaks
encode_list_files(
    experiment_accession="ENCSR...",
    file_format="bed",
    output_type="IDR thresholded peaks",
    assembly="GRCh38"
)

# DNase-seq — may use different output types
encode_list_files(
    experiment_accession="ENCSR...",
    file_format="bed",
    output_type="peaks",
    assembly="GRCh38"
)

Prefer `preferred_default=True` files.

encode_download_files(
    file_accessions=["ENCFF...", ...],
    download_dir="/path/to/data/accessibility",
    organize_by="flat"
)

Validate the downloaded files before filtering. Pass `--assay` so the peak-width and Tn5 checks match the assay; gzipped inputs are read directly.

python3 scripts/validate_peaks.py sample.narrowPeak [--assay atac|dnase|unknown] [--blacklist hg38-blacklist.v2.bed]

Step 4: Per-Sample Noise Filtering

4a. ENCODE Blocklist Filtering (Amemiya et al. 2019)

# Download from: https://github.com/Boyle-Lab/Blacklist/blob/master/lists/hg38-blacklist.v2.bed.gz
gunzip -k hg38-blacklist.v2.bed.gz
bedtools intersect -a sample.narrowPeak -b hg38-blacklist.v2.bed -v > sample.filtered.narrowPeak

4b. SignalValue Filtering (Perna et al. 2024)

Same logic as histone aggregation — filter per-sample to top 75% by signalValue (column 7):

# Per-sample: remove bottom 25% by signalValue (true distribution quantile)
TOTAL=$(wc -l < sample.filtered.narrowPeak)
LINE_25=$(echo "$TOTAL" | awk '{printf "%d