atac-seq-tn5-bias-correction

ATAC-seq Tn5 Insertion Bias Correction

Summary

Correct sequence-dependent Tn5 transposase insertion biases in ATAC-seq BAM files to produce unbiased genome-wide cutsite signals, enabling accurate downstream footprint detection and transcription factor occupancy analysis.

When to use

You have raw ATAC-seq BAM files and need to perform footprinting analysis to detect transcription factor binding through Tn5 insertion patterns. Apply this skill before calculating footprint scores or running differential binding detection, as uncorrected bias will obscure true protein-bound footprints and inflate false positives.

When NOT to use

• Input BAM file is from a non-ATAC assay (e.g., ChIP-seq, DNase-seq) where Tn5 bias is not the dominant systematic error
• You only have already-normalized ATAC-seq counts per peak and no access to the original BAM files with per-base cutsites
• Peak annotations are unavailable or of very poor quality, making it impossible to define the chromatin accessibility context for bias estimation

Inputs

• ATAC-seq BAM file (aligned reads with cutsites)
• Reference genome FASTA file
• Peak annotations BED file (chromatin accessibility regions)

Outputs

• Uncorrected cutsite bigWig file
• Tn5 insertion bias bigWig file
• Expected signal bigWig file
• Bias-corrected cutsite bigWig file
• ATACorrect diagnostic PDF report

How to apply

Run TOBIAS ATACorrect on your BAM file, providing the aligned reads, reference genome in FASTA format, and peak annotations in BED format. The tool models the sequence preference of Tn5 transposase (which shows directional insertion bias at certain DNA sequences) and generates a bias track, expected signal track, and corrected cutsite signal. The corrected bigWig output represents the true Tn5 insertion pattern independent of sequence context, enabling reliable identification of the characteristic insertion depletion around protein-bound sites (footprints). Verify correction quality by visually inspecting the generated PDF report comparing uncorrected vs. corrected signals.