Fragment length estimation from read pairs

fragment-length-estimation-from-read-pairs

Summary

Estimate the average sequencing fragment length from paired-end ChIP-Seq reads using cross-correlation analysis of read pair coordinates. This is a critical preliminary step that informs downstream pileup extension and peak calling sensitivity in ChIP-Seq analysis.

When to use

Apply this skill when you have paired-end ChIP-Seq data (BEDPE format) and need to determine the empirical fragment length (insertion length) before peak calling. This is mandatory for paired-end ChIP-Seq workflows to ensure MACS3 correctly extends coverage tracks and calculates local bias; it is especially important when the biological fragment length is unknown or expected to deviate significantly from default assumptions.

When NOT to use

• Input is single-end ChIP-Seq data (SAM, BAM, or BED format without read-pair information); fragment length estimation requires paired-end coordinate information.
• Fragment length has already been determined by external methods (e.g., Illumina fragment analyzer) and is being supplied manually via --extsize parameter; predictd estimation is redundant.
• Sample is a control/input library rather than the ChIP library; the article states predictd should be applied only to ChIP data, not controls.

Inputs

• BEDPE file (paired-end ChIP sample, e.g., CTCF_PE_ChIP_chr22_50k.bedpe.gz)

Outputs

• Fragment length estimate (scalar, reported in base pairs)
• Cross-correlation profile (internal, used to determine d)

How to apply

Run macs3 predictd with the -f BEDPE flag on the paired-end ChIP sample to estimate the average insertion length via cross-correlation of forward and reverse reads. The subcommand analyzes the distribution of distances between paired reads and outputs a single fragment length estimate (typically reported in base pairs, e.g., ~253 bp for the CTCF_PE_ChIP_chr22_50k.bedpe.gz dataset). Record this estimated fragment length for use in subsequent macs3 callpeak and macs3 pileup steps. The estimated value should be validated against known biology (e.g., typical ChIP sonication fragment sizes) and inspected for outliers that suggest read-pair artifacts or misalignment. When fragment length is successfully estimated, proceed to peak calling without manually specifying --extsize in callpeak mode, allowing the paired-end mode to automatically handle fragment extent during pileup.

Related tools

• macs3 predictd (Estimate average insertion length by cross-correlation analysis of paired-end read positions) — https://github.com/macs3-project/MACS
• macs3 callpeak (Peak calling with paired-end mode that consumes the estimated fragment length for automatic pileup extension) — https://github.com/macs3-project/MACS
• MACS3 (Umbrella toolset providing paired-end ChIP-Seq analysis subcommands including predictd) — https://github.com/macs3-project/MACS

Examples

macs3 predictd -i CTCF_PE_ChIP_chr22_50k.bedpe.gz -f BEDPE

Evaluation signals

• Fragment length estimate is a positive integer within the expected range for sonicated DNA (typically 150–500 bp); check against known sonication target size or prior ChIP-Seq experiments.
• Cross-correlation profile converges to a single clear peak (not bimodal or noisy); inspect diagnostic plots if available.
• Subsequent macs3 callpeak with paired-end mode and no manual --extsize produces narrowPeak output with valid peak coordinates and summit positions.
• Genome background and local bias tracks calculated in downstream steps reflect the estimated fragment length (e.g., d background calculated using the reported d value).
• Peak coordinates and fold-enrichment metrics are stable when re-run on the same data; instability suggests unreliable fragment length estimation.

Limitations

• Cross-correlation method requires sufficient paired-end read depth; sparse or low-quality read pairs may produce noisy or unreliable estimates.
• Estimate is global across the genome; local variation in fragment length due to chromatin structure or PCR bias is not captured.
• Method assumes reads are properly paired; misaligned or corrupted BEDPE files may yield erroneous estimates.
• Single estimate may not be appropriate for mixed-fragment-length libraries; predictd cannot segment fragments by source or PCR cycle.
• The article does not specify expected error bounds or confidence intervals for the fragment length estimate; no quantified uncertainty measure is documented.

Evidence

• [methods] Decide the fragment length d: "This is a crucial step for analyzing ChIP-Seq with MACS3, as well as other types of data"
• [methods] macs3 predictd subcommand: "This can also be accomplished using the predictd subcommand, which we need to apply only to ChIP data"
• [methods] BEDPE input format for paired-end data: "if you run this on the CTCF_PE_ChIP_chr22_50k.bedpe.gz file in the test directory"
• [methods] Fragment length estimate use in callpeak: "Run macs3 callpeak with -f BEDPE flag on the paired-end ChIP sample without specifying --extsize (pileup automatically handles paired-end fragment extent)"
• [methods] Expected fragment length magnitude: "to estimate average insertion length (~253 bp) when processing paired-end ChIP-Seq data in the CTCF_PE_ChIP_chr22_50k.bedpe.gz file"