Explainable Multilayer Graph Neural Network for cancer gene prediction.

Area of Science:

• Computational biology
• Cancer genomics
• Network science

Background:

• Identifying cancer genes is crucial but challenging in cancer genomics.
• Existing methods like deep graph neural networks struggle with multilayered gene interactions and lack explainability.
• Single biological networks fail to capture the complexity of tumorigenesis, leading to inconsistent predictions.

Purpose of the Study:

• To develop an Explainable Multilayer Graph Neural Network (EMGNN) for accurate cancer gene identification.
• To leverage multiple gene-gene interaction networks and pan-cancer multi-omics data.
• To provide explainable predictions and biological insights for identified cancer genes.

Main Methods:

• Utilized a multilayered graph neural network architecture (EMGNN).
• Integrated multiple gene-gene interaction networks and pan-cancer multi-omics data.
• Incorporated model-level feature importance and molecular-level gene set enrichment analysis for explainability.

Main Results:

• EMGNN achieved an average 7.15% improvement in area under the precision-recall curve compared to state-of-the-art methods.
• Successfully prioritized newly predicted cancer genes with conflicting single-network predictions.
• Provided valuable biological insights through explainable AI techniques.

Conclusions:

• EMGNN offers a novel graph learning paradigm for cancer gene identification.
• The method effectively models multilayered topological gene relationships.
• EMGNN serves as a valuable tool for advancing cancer genomics research.