Predicting enhancer-gene links from single-cell multi-omics data by integrating prior Hi-C information

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Enhancers regulate gene expression from distal genomic locations.
• Inferring enhancer-gene links is challenging with single-cell ATAC/RNA-seq alone due to lack of chromatin conformation data.

Purpose of the Study:

• To develop a machine learning method (SCEG-HiC) for decoding enhancer-gene links from single-cell multi-omics data.
• To integrate bulk average Hi-C data as prior knowledge to improve prediction accuracy.

Main Methods:

• Developed SCEG-HiC, a machine learning method using weighted graphical lasso.
• Integrated bulk average Hi-C data with single-cell ATAC/RNA-seq or scATAC-only data.
• Evaluated performance across 10 human and mouse single-cell multi-omics datasets.

Main Results:

• SCEG-HiC outperforms existing single-cell models in predicting enhancer-gene links.
• The method retains context-specific correlations and discovers biologically relevant links.
• Successfully applied to COVID-19 datasets to reconstruct gene regulatory networks and identify noncoding variant associations.

Conclusions:

• SCEG-HiC reliably decodes enhancer-gene links from single-cell multi-omics data.
• The method enhances understanding of gene regulatory networks in disease.
• SCEG-HiC is an open-source R package for broad use in regulatory genomics.