Reference Panel Guided 3D Genome Data Analysis

Abstract

The widespread usage of Hi-C has revealed the hierarchical structures of the genome, thereby deepening our understanding of the organization and function of 3D genomes. However, analyzing Hi-C data remains a challenging task, mainly due to the sequencing coverage of data produced in most Hi-C experiments is insufficient.
In this project, we proposed a reference panel enabled framework to tackle the data insufficiency issue in Hi-C data analysis. This pioneering approach represents the first instance of harnessing the vast amount of existing Hi-C datasets while analyzing a given study Hi-C dataset. Within this framework, we developed three applications to enhance a Hi-C contact map, annotate chromatin loops, and identify nested topologically associating domains (TADs) from insufficiently sequenced Hi-C data. Algorithms developed in this thesis leverage ideas from attention mechanisms, representation learning, dynamic programming, and non-parametric statistics. The introduction of a panel of reference Hi-C samples significantly improved prediction accuracy across three diverse Hi-C data analysis tasks under a wide spectrum of benchmarking scenarios. Applying our tools to Hi-C data from various cells deepened our understanding of the formation of TADs and chromatin loops, unraveling key insights into these essential genomic features.