Unsupervised embedding of single-cell Hi-C data.

Area of Science:

• Genomics
• Molecular Biology
• Computational Biology

Background:

• Single-cell Hi-C (scHi-C) enables the study of 3D genome organization at the individual cell level.
• Understanding cell-to-cell variation in 3D DNA architecture is crucial for developmental and cell-cycle studies.
• Existing Hi-C data analysis methods may not be directly applicable to the unique characteristics of scHi-C data.

Purpose of the Study:

• To evaluate the applicability of established bulk Hi-C data analysis methods to scHi-C data.
• To identify robust computational approaches for analyzing scHi-C datasets.
• To assess the performance of different methods in capturing 3D genome structure variations in single cells.

Main Methods:

• Application of established bulk Hi-C analysis tools to scHi-C datasets.
• Utilizing unsupervised embedding techniques, including multidimensional scaling (MDS).
• Comparative analysis of HiCRep and other methods against a previously used scHi-C technique.

Main Results:

• The HiCRep method, combined with MDS, significantly outperformed three other tested methods.
• The HiCRep/MDS approach demonstrated robustness even with very low per-cell sequencing depth.
• Combining high- and low-coverage cells improved the robustness of the HiCRep/MDS method.
• The method successfully enabled joint embedding of cells from multiple independent scHi-C datasets.

Conclusions:

• The HiCRep/MDS method provides a robust and effective strategy for analyzing scHi-C data.
• This approach is valuable for interrogating 3D genome architecture variability across cells and datasets.
• The findings offer a reliable computational tool for advancing single-cell epigenomics research.