Integrative interpretable learning reveals shared patterns of epitranscriptomic regulation across multiple cancer...

Area of Science:

• Epitranscriptomics
• Cancer Biology
• Genomics

Background:

• Cancer arises from dysregulated cell processes, with N4-acetylcytidine (ac⁴C) acetylation linked to metastasis and tumor progression.
• The shared epitranscriptomic regulatory patterns and interconnected networks across diverse cancer types remain largely unexplored.

Purpose of the Study:

• To develop the first pan-cancer model for ac⁴C epitranscriptome analysis.
• To identify shared epitranscriptomic patterns and regulatory networks of ac⁴C across various cancer types.
• To discover novel ac⁴C-mediated genes associated with cancer prognosis.

Main Methods:

• Utilized 88 ac⁴C epitranscriptome datasets from multiple cancer types and normal tissues.
• Developed a pan-cancer model integrating sequence and genome-derived knowledge using a deep learning transformer architecture.
• Performed interpretable analysis to uncover shared epitranscriptomic patterns and gene associations.

Main Results:

• Uncovered shared epitranscriptomic patterns of dysregulated ac⁴C across cancers, particularly in low GC-content 3'UTR regions and internal 3'UTR splicing.
• Identified candidate ac⁴C-mediated genes involved in cancer epitranscriptomic regulation.
• Discovered three novel ac⁴C-mediated genes (SMARCD1, SENP5, RNF207) associated with poor clinical prognosis across cancer types.

Conclusions:

• Emphasized the significance of comprehensive ac⁴C epitranscriptome characterization in the pan-cancer landscape.
• Highlighted the potential of ac⁴C epitranscriptome insights for developing RNA modification-based biomarkers.
• Suggested implications for novel therapeutic strategies targeting RNA modifications in cancer.