Abstract
In the past eleven years, Hi-C techniques have been widely used in exploring three-dimensional (3D) chromosomal conformation. Raw Hi-C data are usually processed through a series of steps, such as resolution enhancement and normalization, for being used on capturing significant structural signatures, such as 3D structure reconstruction, calling topologically associating domains (TADs), and chromatin loops. We developed a new computational method named HiCNN that used a 54-layer deep convolutional neural network to enhance the resolutions of Hi-C data. The network contains both global and local residual learning with multiple speedup techniques included for fast convergence. The evaluation results show that HiCNN consistently outperforms the state-of-the-art method HiCPlus. We further present a deep-learning package named HiCNN2, an improved version of HiCNN. To normalize single-cell Hi-C and genome architecture mapping (GAM) data, we introduced two methods (scHiCNorm and normGAM). We developed a new method to infer this converting parameter and pairwise Euclidean distances based on the topology of the Hi-C complex network (HiCNet). The inferred distances were modeled by clustering coefficient and multiple other types of constraints. For exploring structural properties of TADs, we designed and benchmarked three metrics for capturing the 3D and two-dimensional (2D) structural signatures of TADs. We built an online knowledge base, TADKB, integrating knowledge for TADs in eleven cell types of human and mouse. TADKB provides predicted 3D structures of chromosomes and TADs and detailed annotations about protein-coding genes and long non-coding RNAs (lncRNAs) existent in each TAD. For 3D structure reconstruction and TAD detection in single-cell Hi-C data, we present scHiCEmbed, an integrated framework based on graph auto-encoder. It can embed each bin in a given dimensional space for modeling 3D organizations of the whole genome in individual cells and classifying bins that are spatial proximity to each other for calling TADs