GEDFN

GEDFN integrates graph embedding with a deep feedforward network to identify microbial biomarkers from microbiome data for microbiome-wide association studies.

Key Features:

• Sparse Microbial Interaction Network Construction: Constructs sparse microbial interaction networks to capture interactions within high-dimensional, noisy microbiome data.
• Graph Embedding Deep Learning Integration: Integrates graph embedding techniques with a deep feedforward network to embed interaction networks into a predictive model.
• Feature Selection Capability: Performs feature selection based on graph embedding to identify and validate biologically relevant microbial markers.
• Interpretability and Flexibility: Combines graph-based representations with deep models to improve interpretability of selected microbial features.

Scientific Applications:

• Microbiome-wide association studies: Links microbial features to host phenotypes within microbiome-wide association studies.
• Disease biomarker discovery: Facilitates discovery of microbial biomarkers for disease diagnosis.
• Human–microbe interaction analysis: Enables analysis of human–microbe interactions by capturing complex microbial relationships.

Methodology:

Constructs various microbial interaction networks using different approaches, embeds these networks via graph embedding, and integrates the embeddings into a deep feedforward network for feature selection and prediction.