Multiscale and integrative single-cell Hi-C analysis with Higashi

Abstract

AbstractThe advent of single-cell Hi-C (scHi-C) technologies offers an unprecedented opportunity to unveil cell-to-cell variability of 3D genome organization. However, the development of computational methods that can effectively enhance scHi-C data quality and extract 3D genome features in single cells remains a major challenge. Here, we report Higashi, a new algorithm that achieves state-of-the-art analysis of scHi-C data based on hypergraph representation learning. Extensive evaluations demonstrate that Higashi significantly outperforms existing methods for embedding and imputation of scHi-C data. Higashi is uniquely able to identify multiscale 3D genome features (such as compartmentalization and TAD-like domain boundaries) in single cells, allowing markedly refined delineation of cell-to-cell variability of 3D genome features. By applying to a scHi-C dataset from human prefrontal cortex, Higashi reveals complex cell types as well as new connections between 3D genome features and cell type-specific gene regulation. Higashi provides an end-to-end solution to scHi-C data analysis and is applicable to studying single-cell 3D genomes in a wide range of biological contexts.