An Attention-Based Deep Neural Network Model to Detect Cis-Regulatory Elements at the Single-Cell Level From...

Area of Science:

• Genomics and Bioinformatics
• Single-cell Biology
• Computational Biology

Background:

• Cis-regulatory elements (cREs) are vital for gene expression control, cell differentiation, and state transitions.
• Existing methods for analyzing cRE activity provide population averages, masking crucial cell-specific variations.
• Single-cell resolution is essential for understanding heterogeneous cell state dynamics.

Purpose of the Study:

• To develop a novel computational framework for detecting cREs and their activities at the single-cell level.
• To integrate DNA sequence, genomic distance, and single-cell multi-omics data for enhanced cRE detection.
• To improve the accuracy of cell clustering and differentiation based on predicted cRE activities.

Main Methods:

• An attention-based deep neural network framework was designed.
• The model integrates DNA sequences, genomic distances, and single-cell multi-omics data.
• Performance was evaluated on human peripheral blood mononuclear cells and glioma patient data.

Main Results:

• The proposed model demonstrated superior accuracy in identifying cREs in single-cell multi-omics data compared to existing methods.
• Predicted cRE activities enabled more precise cell clustering and finer differentiation of cell states.
• The model identified tumor-specific SOX2 activity and heterogeneous ZEB1 activation in glioma data, which are challenging for conventional approaches.

Conclusions:

• The developed deep learning framework effectively detects cis-regulatory element regulation at the single-cell level.
• This approach enhances the understanding of cellular heterogeneity and gene regulatory mechanisms.
• The model holds potential for uncovering drug resistance mechanisms within specific cell sub-populations.