local-background-bias-estimation

local-background-bias-estimation

Summary

Construct a multi-scale local background noise track from control ChIP-Seq data by computing fragment-length (d), short-local (slocal, 1kb), and long-local (llocal, 10kb) pileups, normalizing each scale relative to fragment-length baseline, and combining them via maximum operation to capture spatially varying sequencing bias. This track is then scaled to match ChIP sequencing depth and used as the null hypothesis for peak enrichment testing.

When to use

Apply this skill when you have paired ChIP and control BED/BEDPE files and need to account for local sequencing bias before peak calling. Use it specifically when control signal varies across genomic regions at multiple spatial scales (e.g., open chromatin bias at 1kb and 10kb windows differs from global background), or when simple global normalization would mask true enrichment in high-background regions.

When NOT to use

• Input control file is already normalized or has been processed through another bias-correction pipeline; applying local-lambda estimation a second time risks double-correction.
• Sequencing depth is extremely low (< 1M reads in control) or extremely high (> 1B reads), making multi-scale window estimates unstable or memory-prohibitive.
• Analysis goal is to call broad peaks (histone marks with diffuse enrichment); use macs3 bdgbroadcall instead, which applies its own broad-scale background model.

Inputs

• Filtered control BED file (after duplicate removal via macs3 filterdup)
• Predicted fragment length d (from macs3 predictd on ChIP sample)
• Control read count (final count after duplicate filtering)
• ChIP read count (final count after duplicate filtering)
• Genome size (in base pairs, e.g., 2.7 billion for human haploid)

Outputs

• Local lambda BEDGRAPH file (scaled to ChIP sequencing depth)
• Intermediate d, slocal, llocal BEDGRAPH files (if retained for inspection)
• Combined maximum background BEDGRAPH (before depth scaling)