DIRECT-NET: An efficient method to discover cis-regulatory elements and construct regulatory networks from...

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Single-cell multiomics data offers insights into cell identity.
• Dissecting cis-regulatory element (CRE)-to-gene relationships at single-cell resolution is challenging.

Purpose of the Study:

• Introduce DIRECT-NET, a machine learning method for CRE-gene relationship inference.
• Evaluate DIRECT-NET's performance against existing methods.
• Enable prediction of transcriptional regulation from single-cell data.

Main Methods:

• Developed DIRECT-NET, a gradient boosting-based machine learning model.
• Utilized parallel single-cell gene expression and chromatin accessibility data, or chromatin accessibility data alone.
• Validated DIRECT-NET predictions using independent functional genomics data.

Main Results:

• DIRECT-NET accurately identifies genome-wide CREs and their target genes.
• Significantly improved accuracy in inferring CRE-to-gene relationships compared to current methods.
• Revealed cell subpopulation-specific and dynamic regulatory linkages.

Conclusions:

• DIRECT-NET is an efficient tool for CRE-gene relationship prediction.
• Facilitates the study of transcriptional regulation using single-cell multiomics.
• Advances our understanding of cell identity mechanisms.