Graph Neural Networks With Multiple Prior Knowledge for Multi-Omics Data Analysis.

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Precision medicine relies on analyzing multi-omics data.
• Existing graph neural network (GNN) methods for multi-omics data analysis have not fully utilized multiple sources of prior biological knowledge simultaneously.
• Gene-gene interaction networks are examples of graph-based biological knowledge.

Purpose of the Study:

• To propose a novel multi-omics data analysis framework, MPK-GNN, that integrates multiple prior biological knowledge graphs.
• To address the limitation of existing methods in simultaneously exploiting knowledge from multiple graph sources.
• To enhance multi-omics learning for precision medicine applications.

Main Methods:

• MPK-GNN incorporates multiple prior graphs into a GNN framework.
• The method includes a feature-level learning module to aggregate information from prior graphs.
• A projection module optimizes agreement among prior networks using contrastive loss, a sample-level module learns global representations, and a task-specific module allows flexible extension for downstream analyses.

Main Results:

• MPK-GNN was evaluated on cancer molecular subtype classification.
• Experimental results demonstrated that MPK-GNN outperforms state-of-the-art algorithms.
• The proposed method showed superiority over existing multi-view learning and multi-omics integrative approaches.

Conclusions:

• MPK-GNN represents the first framework to integrate multiple prior graphs into multi-omics data analysis.
• The framework effectively enhances multi-omics learning by leveraging diverse biological knowledge.
• MPK-GNN shows significant potential for advancing precision medicine through improved molecular subtype classification.