poisson-test-statistical-scoring

Poisson Test Statistical Scoring

Summary

Apply Poisson or quasi-Poisson statistical models to compare ChIP-Seq signal against local background noise, generating p-value or q-value scores for each genomic position. This skill transforms raw coverage comparisons into statistical significance estimates suitable for peak calling.

When to use

After generating ChIP pileup and local lambda (background) BEDGRAPH tracks with matched sequencing depth, use this skill to assign statistical significance scores to each genomic region. Specifically, when you have comparable ChIP and control coverage tracks and need to identify enriched regions above baseline noise — typical in narrow peak calling workflows where a single comparison model (qpois or ppois) must be applied genome-wide.

When NOT to use

• Input backgrounds are not normalized to the same sequencing depth as ChIP — local lambda must be scaled by the ratio (final_ChIP_reads / control_reads) before comparison.
• ChIP and control samples have vastly different library sizes without depth normalization; the statistical model assumes comparable sequencing effort.
• Local background was not properly constructed from d, slocal (1kb), and llocal (10kb) windows; misspecified background invalidates the significance test.

Inputs

• ChIP pileup BEDGRAPH (coverage track extended to fragment length d)
• Local lambda background BEDGRAPH (scaled to ChIP sequencing depth, combining d/slocal/llocal maximum with genome background)

Outputs

• q-value BEDGRAPH (log10-transformed quasi-Poisson scores, base-pair resolution)
• p-value BEDGRAPH (log10-transformed Poisson scores, base-pair resolution)

How to apply

Use macs3 bdgcmp with the -m qpois flag to apply a quasi-Poisson test comparing ChIP pileup BEDGRAPH against the scaled local lambda background BEDGRAPH. The quasi-Poisson model accounts for overdispersion in read counts and outputs a q-value BEDGRAPH; alternatively, use -m ppois for p-value scoring if Poisson assumptions hold more strictly. The resulting score track contains log10-transformed statistical significance at base-pair resolution. Choose qpois (default, more robust) when read count variance exceeds the mean; choose ppois when variance equals mean. The output BEDGRAPH is then thresholded using macs3 bdgpeakcall with a cutoff (e.g., -c 1.301 for q-value ≤ 0.05) to call peaks.