Topologically associating domains of chromatin on single-cell Hi-C data: a survey of bioinformatic tools and applications in the light of artificial intelligence.

Abstract

Topologically associating domains (TADs) uncovered on bulk Hi-C data are regarded as fundamental building blocks of a three-dimensional genome, and they are believed to effectively participate in the regulatory programs of gene expression. The computational analysis of TADs on single-cell Hi-C (scHi-C) data in the era of single-cell transcriptomics has received continuous attention since it may provide information beyond that on bulk Hi-C data. Unfortunately, the contact matrix for a single cell is ultra-sparse due to the low sequencing depth. Coupled with noises, artifacts, and dropout events from experiments, as well as cell heterogeneity caused by the cell cycle and transcription status, the computational analysis of TAD structures at the single-cell level has encountered some challenges not encountered at the bulk level. Herein, conduct a survey of bioinformatic tools and applications for TAD structures at the single-cell level in the light of artificial intelligence, including imputation of scHi-C data, identification of TAD boundaries and hierarchy, and differential analysis of TAD structures. The categories, characteristics, and evolutions of the latest available methods are summarized, especially the artificial intelligence strategies involved in these issues. This is followed by a discussion on why deep neural networks are attractive when discovering complex patterns from scHi-C data with an enormous number of cells and how it promotes the computational analysis of TADs at the single-cell level. Furthermore, the challenges that may be encountered in the analysis are outlined, and an outlook on the emerging trends in the near future is presented cautiously.