An omics data analysis method based on feature linear relationship and graph convolutional network.

Area of Science:

• Systems biology
• Bioinformatics
• Network medicine

Background:

• Biological networks exhibit modularity, and module dysfunction is linked to disease.
• Current omics data analysis for disease classification often overlooks network interactions, limiting model performance.

Purpose of the Study:

• To develop an omics data analysis method that integrates feature linear relationships and network topology for improved disease classification.
• To identify disease-specific modules and biomarkers from omics data.

Main Methods:

• Proposed LCNet method utilizing feature linear relationships and graph convolutional networks (GCN).
• Constructed differential linear relation networks to capture physiological and pathological changes.
• Employed a greedy strategy for module searching and defined personalized sub-graphs for GCN classification.

Main Results:

• LCNet demonstrated superior classification performance on public datasets.
• Identified relevant metabolites and pathways for Breast Cancer using metabolic data.
• The method effectively utilizes both node and topological information within identified modules.

Conclusions:

• LCNet provides a novel approach for disease classification by integrating molecular feature relationships and network structures.
• The method successfully identifies module biomarkers and enhances diagnostic capabilities.
• LCNet offers a new paradigm for leveraging omics data and network information in disease research.