Robust single-cell Hi-C clustering by convolution- and random-walk-based imputation.

Area of Science:

• Genomics
• Computational Biology
• Cell Biology

Background:

• Three-dimensional genome structure is crucial for gene regulation and cellular functions.
• Single-cell genome architecture analysis uses methods like Hi-C, but requires computational approaches for sparse data.
• Existing methods struggle with accurate and efficient clustering of heterogeneous single-cell Hi-C data.

Purpose of the Study:

• To develop a novel computational algorithm, scHiCluster, for accurate clustering of single-cell Hi-C data.
• To enable the study of chromosome structure variations across different cell types.
• To improve the identification and comparison of 3D genome structures at the single-cell level.

Main Methods:

• scHiCluster employs imputation via linear convolution and random walk for Hi-C contact matrices.
• The algorithm was benchmarked using simulated and real single-cell Hi-C datasets.
• Performance was compared against existing clustering methods, particularly for low-coverage data.

Main Results:

• scHiCluster significantly enhances clustering accuracy on sparse, low-coverage single-cell Hi-C data.
• The method enables the identification of topologically associating domain (TAD)-like structures (TLSs) within individual cells post-imputation.
• Consensus boundaries of identified TLSs show enrichment at TAD boundaries found in bulk Hi-C samples.

Conclusions:

• scHiCluster provides an accurate and efficient solution for clustering single-cell Hi-C data.
• The algorithm facilitates the visualization and comparison of 3D genome organization across single cells.
• This advancement aids in understanding cell-type-specific genome architecture and its regulatory implications.