HGCPep: Hypergraph Deep Learning Identifies Cancer-associated Non-coding Peptides.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Non-coding peptides (ncPEPs) encoded by non-coding RNAs (ncRNAs) are emerging as critical regulators and biomarkers in cancer.
• Current computational methods for ncPEP identification primarily rely on sequence analysis, neglecting the shared transcriptional origin of peptides from a single ncRNA.

Purpose of the Study:

• To develop a novel deep learning framework, HGCPep, that models the intrinsic relationships between ncRNAs and their encoded ncPEPs.
• To improve the systematic identification of cancer-associated ncPEPs by incorporating transcriptional context.

Main Methods:

• Developed HGCPep, a deep learning framework utilizing hypergraphs to represent ncRNAs and their encoded peptides.
• Integrated a hypergraph neural network with a convolutional neural network to enrich peptide feature representations with transcriptional context.
• Applied dimensionality reduction to learned embeddings to analyze ncPEP clustering by cancer type.

Main Results:

• HGCPep outperforms state-of-the-art methods in identifying cancer-associated ncPEPs.
• Learned embeddings from HGCPep reveal distinct clustering of ncPEPs based on cancer type, indicating effective deciphering of biological associations.
• The framework provides a powerful tool for discovering novel therapeutic targets in oncology.

Conclusions:

• HGCPep offers a novel and effective method for ncPEP analytics by leveraging hypergraph modeling.
• The framework enhances the identification of cancer-associated ncPEPs and aids in understanding complex biological associations.
• HGCPep represents a significant advancement in the discovery of novel therapeutic targets for cancer immunotherapy.