Prior knowledge-guided multilevel graph neural network for tumor risk prediction and interpretation via multi-omics...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration offers insights into complex disease mechanisms but faces challenges like high dimensionality and limited samples.
• Prior biological knowledge, including gene regulatory networks and pathways, contains valuable gene interaction and functional information.
• Existing methods struggle to fully leverage the complementary nature of multi-omics data and prior knowledge.

Purpose of the Study:

• To develop a novel deep learning approach for effective multi-omics data integration.
• To incorporate prior biological knowledge (gene regulatory networks, pathways) into the analysis.
• To improve the accuracy of disease survival risk prediction using integrated data.

Main Methods:

• Proposed a Multilevel-graph neural network (GNN), a hierarchical deep learning algorithm.
• Sequentially leveraged multi-omics data, gene regulatory networks, and pathway information for feature extraction.
• Employed GNN-Explainer and IGscore for interpreting neural network results at gene and pathway levels.

Main Results:

• The proposed Multilevel-GNN achieved superior accuracy in survival risk prediction compared to existing methods.
• Identified key genes (e.g., SEC61G, CYP27B1) and pathways nonlinearly associated with tumor pathogenesis.
• Prioritized factors demonstrated strong associations with disease in survival analyses, with many previously reported in literature.

Conclusions:

• The Multilevel-GNN effectively integrates multi-omics data and prior biological knowledge for enhanced disease survival risk prediction.
• The approach provides a powerful tool for uncovering complex molecular mechanisms underlying diseases.
• Interpretation algorithms successfully identified biologically relevant genes and pathways associated with disease prognosis.