Accurate chromatin marks peak calling with Omnipeak.

Chromatin immunoprecipitation sequencing (ChIP-seq) is a widely used technique for identifying transcription-factor binding and histone modifications across the genome. Peaks in ChIP-seq data vary in length depending on the biological context, from narrow transcription factor binding sites to broad histone modification domains. However, commonly used peak-calling tools are tailored to the specific types of data and struggle to consistently handle various peak lengths, datasets of varying quality, and missing control tracks-common issues in comparative or meta-analyses. These limitations are addressed with Omnipeak, a universal unsupervised peak-calling algorithm based on a constrained three-state hidden Markov model. Omnipeak accurately models global genomic read coverage, capturing structure patterns in the data of all length scales and variable quality. We benchmarked Omnipeak versus eight different peak calling methods using over 550 public and 300 synthetic datasets, including conventional, ultra-low-input ChIP-seq, and ATAC-seq. Omnipeak produced consistent peaks across narrow, broad, and variable mark lengths, with the best agreement between replicates and robustness against noise and lack of control tracks. Together with a variety of supported input formats and peak calling capabilities within the genome browser, Omnipeak is well-positioned for processing various ChIP-seq and ATAC-seq datasets.