fragment-length-prediction-and-extension

fragment-length-prediction-and-extension

Summary

Estimate the DNA fragment length (d) from ChIP-Seq data using cross-correlation analysis, then extend aligned reads to this length to generate accurate ChIP coverage tracks. This step is critical for converting point-wise read alignments into fragment-level signal representation.

When to use

After filtering duplicate reads from ChIP-Seq data but before generating pileup coverage tracks. Apply this skill when you have aligned ChIP reads in BED format and need to construct coverage BEDGRAPH files that reflect the actual DNA fragment distribution rather than single-end read positions.

When NOT to use

• Input reads are already paired-end (BEDPE format) — use the observed fragment length distribution from BEDPE instead of re-predicting.
• Fragment length is known a priori from sequencing metadata — skip prediction and use the known value directly with --extsize.
• Control sample is being processed — predictd should only be applied to ChIP data, not input/control data.

Inputs

• Filtered ChIP-Seq reads in BED format (e.g., CTCF_ChIP_200K.bed.gz)
• Genome size specification (e.g., 'hs' for human, or explicit bp count)

Outputs

• Predicted fragment length scalar d (integer, in base pairs)
• ChIP coverage BEDGRAPH track (genomic coordinates × extended fragment count)

How to apply

First, run macs3 predictd on the filtered ChIP sample using parameters -g hs (human genome size) and -m 5 50 (search fragment length range 5–50 bp) to estimate the dominant fragment length d from cross-correlation peaks. The tool analyzes the distribution of reads across the genome and returns a single scalar d value. Next, use macs3 pileup with the --extsize parameter set to d to extend each filtered ChIP read bidirectionally (or in a strand-aware manner) to the predicted fragment length. This produces a BEDGRAPH file where each genomic position reflects the number of extended fragments covering it, not raw reads. The predicted fragment length d should typically fall within the sequencing protocol's expected range (e.g., 150–200 bp for typical ChIP-Seq); anomalous values (< 50 bp or > 500 bp) warrant re-examination of sequencing quality and read filtering.