DeepECC: a deep learning framework for genome-wide identification and analysis of human cancer eccDNAs

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Extrachromosomal circular DNAs (eccDNAs) are prevalent in eukaryotes, influencing gene regulation and cancer.
• Current experimental methods for eccDNA detection are limited by cost and completeness.
• Computational approaches are needed to address eccDNA heterogeneity and biogenesis.

Purpose of the Study:

• To develop a deep learning framework, DeepECC, for accurate and efficient eccDNA identification.
• To analyze eccDNA sequence features and genomic distributions to understand their regulatory functions.
• To provide a tool for visualizing and querying cancer-related eccDNAs.

Main Methods:

• A two-stage deep learning strategy was employed to model sequence context around eccDNA breakpoints.
• DeepECC was trained and validated on multi-species datasets (human, mouse, gallus).
• Genome-wide scanning and motif analysis were performed on predicted eccDNAs.

Main Results:

• DeepECC effectively captures conserved breakpoint features, favoring GC-rich and transcriptionally active regions.
• Human cancer eccDNAs show non-uniform distributions, enriched near enhancers, eQTLs, and CTCF sites.
• Motif analysis suggests eccDNA involvement in ribosomal activity, translational regulation, and DNA damage response.
• Predicted eccDNAs were integrated into the UCSC Genome Browser for visualization and functional interpretation.

Conclusions:

• DeepECC offers a generalizable computational framework for systematic eccDNA discovery.
• eccDNAs play significant regulatory roles in cancer progression.
• The DeepECC tool facilitates experimental research by enabling convenient querying and visualization of cancer-associated eccDNAs.