Graph Embedding Deep Learning Guides Microbial Biomarkers' Identification.

Area of Science:

• Microbiome research
• Bioinformatics
• Computational biology

Background:

• Microbiome-wide association studies (MWAS) aim to identify microbial biomarkers for disease diagnosis.
• Microbiome data is high-dimensional and noisy, posing challenges for traditional machine learning.
• Deep learning models require substantial data and lack interpretability, limiting their application in MWAS.

Purpose of the Study:

• To develop a novel deep learning framework for analyzing complex microbiome data.
• To improve the identification of microbial biomarkers for disease diagnosis.
• To address the limitations of traditional machine learning and deep learning in MWAS.

Main Methods:

• Construction of sparse microbial interaction networks.
• Integration of graph embedding techniques with deep learning models.
• Development of a Graph Embedding Deep Feedforward Network (GEDFN) for feature selection.

Main Results:

• Demonstrated the feasibility of combining microbial graph models with deep learning.
• Successfully applied deep learning and feature selection to microbiome data.
• Validated the biological significance of identified microbial markers.

Conclusions:

• The proposed GEDFN framework effectively analyzes complex microbiome data.
• This approach enhances the identification of microbial biomarkers for disease diagnosis.
• The integration of graph embedding and deep learning offers a promising direction for MWAS.