Peakcalling

Peakcalling is a computational process used in high-throughput sequencing experiments to identify regions of the genome that show a significant enrichment of sequencing reads relative to a background model. In ChIP-seq, ATAC-seq, DNase-seq and similar assays, peaks correspond to protein-DNA binding sites or accessible chromatin, respectively.

Input data typically include aligned reads in BAM format, and often a control sample (input DNA or IgG) to model background noise. Preprocessing may include read filtering, duplicate removal, shifting or adjusting fragment lengths to reflect the experimental protocol (for ATAC-Seq and DNase-Seq, promoting Tn5 cut-site interpretation), and genome-wide mappability considerations.

Peak calling relies on statistical models that compare observed read counts to expected background, accounting for local biases. Common approaches use Poisson or negative binomial tests, sometimes with window-based smoothing. Popular tools include MACS2 (widely used for ChIP-Seq), Genrich (ATAC-Seq and ChIP-Seq), SICER (broad domains), and HOMER. Outputs typically include a set of peaks with genomic coordinates, signal scores, p-values, q-values (false discovery rate), and sometimes summit positions or fold enrichment. Replicate analysis can use IDR to measure reproducibility.

Post-processing may involve annotating peaks relative to genes or regulatory features, motif enrichment analysis, and integration with other data. Limitations include dependence on sequencing depth and quality, choice of control, biases due to genome composition and mappability, and the presence of broad versus narrow peak types. Blacklist regions should be considered. Reproducibility across replicates is commonly assessed with IDR.