Advanced deepTools Platform

A scientific computing skill for analyzing high-throughput sequencing data using deepTools — the suite of Python tools for processing and visualizing next-generation sequencing data, particularly ChIP-seq, ATAC-seq, and RNA-seq signal tracks. Advanced deepTools Platform helps you generate coverage tracks, create heatmaps, and perform quality control on aligned sequencing data.

When to Use This Skill

Choose Advanced deepTools Platform when:

• Converting BAM alignments to bigWig coverage tracks
• Creating heatmaps and profile plots around genomic features
• Performing quality control on ChIP-seq or ATAC-seq data
• Comparing signal intensity across multiple samples or conditions

Consider alternatives when:

• You need peak calling (use MACS2 or MACS3)
• You need differential binding analysis (use DiffBind)
• You need genome assembly (use SPAdes or similar)
• You need variant calling (use GATK or bcftools)

Quick Start

Copy
claude "Generate a coverage bigWig and create a heatmap around TSS regions"

Copy
# Convert BAM to normalized bigWig
bamCoverage \
  --bam sample.bam \
  --outFileName sample.bw \
  --binSize 10 \
  --normalizeUsing RPKM \
  --numberOfProcessors 8 \
  --extendReads 200

# Compute matrix around TSS
computeMatrix reference-point \
  --referencePoint TSS \
  --scoreFileName sample.bw \
  --regionsFileName genes.bed \
  --outFileName matrix.gz \
  --upstream 3000 \
  --downstream 3000 \
  --binSize 50

# Plot heatmap
plotHeatmap \
  --matrixFile matrix.gz \
  --outFileName heatmap.png \
  --colorMap RdYlBu_r \
  --sortUsing mean \
  --missingDataColor white

Core Concepts

deepTools Commands

Command	Purpose	Input → Output
bamCoverage	BAM to bigWig signal track	BAM → bigWig
bamCompare	Log2 ratio of two BAM files	2 BAMs → bigWig
computeMatrix	Extract signal around features	bigWig + BED → matrix
plotHeatmap	Heatmap from signal matrix	matrix → PNG/PDF
plotProfile	Average profile plot	matrix → PNG/PDF
multiBamSummary	Correlation across samples	BAMs → npz matrix
plotCorrelation	Sample correlation heatmap	npz → PNG/PDF
plotFingerprint	ChIP enrichment QC	BAMs → PNG/PDF

Normalization Methods

Copy
# RPKM — Reads Per Kilobase per Million mapped reads
bamCoverage --normalizeUsing RPKM

# CPM — Counts Per Million mapped reads
bamCoverage --normalizeUsing CPM

# BPM — Bins Per Million (like TPM for coverage)
bamCoverage --normalizeUsing BPM

# RPGC — Reads Per Genomic Content (1x normalization)
bamCoverage --normalizeUsing RPGC --effectiveGenomeSize 2913022398

Multi-Sample Comparison

Copy
# Compare ChIP vs Input (log2 ratio)
bamCompare \
  --bamfile1 chip.bam \
  --bamfile2 input.bam \
  --outFileName chip_vs_input_log2.bw \
  --operation log2 \
  --normalizeUsing RPKM

# Sample correlation matrix
multiBamSummary bins \
  --bamfiles sample1.bam sample2.bam sample3.bam \
  --outFileName multisample.npz \
  --binSize 1000

plotCorrelation \
  --corData multisample.npz \
  --plotFile correlation.png \
  --corMethod pearson \
  --whatToPlot heatmap

Configuration

Parameter	Description	Default
binSize	Resolution in base pairs	50
normalizeUsing	Normalization method	None
numberOfProcessors	CPU threads to use	1
extendReads	Extend reads to fragment length	False
effectiveGenomeSize	Mappable genome size (for RPGC)	Organism-specific

Best Practices

1. Always normalize coverage tracks for comparison. Raw coverage depends on sequencing depth. Use RPKM, CPM, or RPGC normalization when comparing samples. For ChIP-seq, use bamCompare with input control for the most meaningful signal.

2. Extend reads to fragment length. For single-end sequencing, use --extendReads 200 (or your estimated fragment length) to reconstruct the full fragment footprint. This produces smoother, more biologically meaningful coverage tracks.

3. Use appropriate bin sizes for your analysis. Smaller bins (10bp) provide higher resolution but larger files and slower processing. For genome-wide heatmaps, 50bp bins are usually sufficient. For narrow peak analysis, use 10bp bins.

4. Run plotFingerprint for ChIP-seq QC. Before downstream analysis, verify ChIP enrichment quality. A good ChIP sample shows a characteristic curve distinct from input — flat curves indicate poor enrichment and unreliable results.

5. Sort heatmaps meaningfully. Use --sortUsing mean or --sortUsing max to order regions by signal intensity. This reveals patterns in the data (e.g., gene clusters with different signal levels) that random ordering obscures.

Common Issues

bigWig file is much larger than expected. Reduce bin size (larger value = smaller file) and ensure you're using normalization (normalized values compress better). Use --skipZeroOverZero to omit regions with no coverage.

Heatmap shows uniform color with no pattern. The signal range may be dominated by outliers. Set --zMin and --zMax manually to clip extreme values, or use --sortUsing mean to reveal signal patterns hidden by default sorting.

bamCoverage is extremely slow. Use --numberOfProcessors 8 (or more) for parallelization. Ensure the BAM file is sorted and indexed (samtools sort + samtools index). Remove duplicate reads before processing to reduce computation.