iHerd: an integrative hierarchical graph representation learning framework to quantify network changes and prioritize...

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Cellular functions rely on complex gene networks, and alterations can lead to disease.
• Identifying critical driver genes and quantifying network changes across conditions is essential for understanding biological processes and disease mechanisms.

Purpose of the Study:

• To introduce iHerd, a hierarchical graph representation learning method for quantifying gene network alterations and prioritizing driver genes.
• To enable the classification of driver genes into early and late divergent genes (EDGs and LDGs) for deeper molecular insights.

Main Methods:

• iHerd employs hierarchical graph coarsening to represent network modules at multiple resolutions.
• It uses efficient graph embedding to learn node representations across all hierarchical levels.
• A graph alignment module projects gene embeddings into a shared latent space to compute a rewiring index for driver gene prioritization.

Main Results:

• iHerd successfully identified novel and known disease risk genes in tumor-normal and cell-type-specific analyses.
• The method effectively classified driver genes into EDGs and LDGs, highlighting genes with significant network changes at various pathway levels.
• Application to single-cell multiome brain data demonstrated iHerd's capability in complex network analysis.

Conclusions:

• iHerd provides an efficient and interpretable approach for analyzing gene regulatory network rewiring.
• The hierarchical learning and classification of driver genes offer unique molecular insights into disease mechanisms.
• The developed method advances the field of systems biology by enabling a nuanced understanding of gene network dynamics.