Towards understanding genome regulation via high-resolution analysis of chromatin accessibility

Thesis

Next generation sequencing has been used to in functional genomics to rep- resent specific aspects of chromatin structure, DNA-protein interactions and the epigenome, enriching the knowledge of the non-coding genome which plays significantly roles in gene regulation but is vaguely understood. Among the assays for functional genomics, Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) is a powerful and popular tool as its simple read out of ‘chromatin accessibility’ provides both the positioning of functional regulatory elements and a general representation of the active genome. Despite its advantages, the analysis of ATAC-seq is challenging and due to the data sparsity and the sub-optimal use of the data, especially the fragment size of the sequencing reads. In this thesis, I address the use of ATAC-seq fragment size in two different ways with respect to different goals: prioritising functional variants and peak calling.

The first goal is achieved through a collaborative work in Avocato, which develops an end-to-end pipeline for single-cell ATAC-seq (scATAC-seq) functional analysis. Avocato employs a simple but useful fragment size filtering strategy, retaining only short fragments from high-quality scATAC-seq data, which successfully prioritises functional SNP variants hidden in the raw signals. Several statistical optimisations in pre-processing stages and a powerful interactive visualisation platform assist Avocato to be a solid and hands-on tool for high- resolution scATAC-seq analysis.

The second goal, peak calling, focuses on both bulk and single-cell ATAC-seq data. I developed EpiCall, a novel ATAC-seq specific peak caller that shows superior performance compared to current popular peak callers. By modeling ATAC-seq fragments of varying sizes differently, EpiCall optimally uses fragment size and coverage information, which is integrated into a mechanistic model for the actual formation of sequencing fragments. I applied EpiCall in various datasets and evaluated it extensively with current peak callers. For bulk ATAC- seq datasets, I demonstrate the precision and recall between peak calls and ENCODE functional annotations, as well as the enrichment of motifs and histone modifications. For scATAC-seq datasets, I show the ability of EpiCall to improve differential motif enrichment across single cells and cell clusters, and in detecting and distinguishing neighborhood peaks sensitively. For both bulk and single-cell datasets, I show the robustness of EpiCall in peak calling down-sampled low-quality data, a significant advantage over current peak callers.

Authors: Dai, L

• DOI: 10.5287/ora-dav75pnkb
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: multi-omics; chromatin accessibility; Hidden Markov Model; ATAC-seq; gene regulation; single-cell sequencing; peak calling
• Subjects: Bioinformatics; Nucleotide sequence--Data processing; Functional genomics