A Lightweight Framework For Chromatin Loop Detection at the Single-Cell Level.

Area of Science:

• Genomics and Bioinformatics
• Molecular and Cell Biology

Background:

• Single-cell Hi-C (scHi-C) enables chromatin organization analysis at the individual cell level.
• scHi-C data is sparse, posing challenges for existing loop-calling methods that require dense contact maps.
• Current methods incur high computational costs and lose single-cell resolution due to data imputation.

Purpose of the Study:

• To develop a computationally efficient and accurate framework for loop calling in sparse scHi-C data.
• To overcome the limitations of existing methods in handling data sparsity and preserving single-cell information.
• To enable the investigation of chromatin looping variability and regulatory roles at the single-cell level.

Main Methods:

• Introduced scGSLoop, a lightweight framework utilizing graph-based deep learning.
• Adapted training and inference strategies to leverage genomic sequence features and 1D positional information.
• Designed the model to exploit data sparsity as an advantage for computational efficiency.

Main Results:

• scGSLoop achieves unprecedented computational efficiency by effectively handling sparse scHi-C data.
• The framework demonstrates higher accuracy in loop prediction compared to existing methods.
• scGSLoop identifies a greater number of potentially regulatory loops and preserves distinct looping information for each cell.

Conclusions:

• scGSLoop sets a new paradigm for scHi-C loop calling, addressing sparsity and computational challenges.
• The method enhances the understanding of chromatin looping variability and regulatory mechanisms across single cells.
• scGSLoop's capabilities can be extended to investigate complex multi-connected genomic hubs.