Human disease prediction from microbiome data by multiple feature fusion and deep learning.

Area of Science:

• Metagenomics
• Microbiome analysis
• Computational biology

Background:

• Human disease prediction from microbiome data is crucial but current methods often lose information by not considering known/unknown microbes or taxonomic relationships.
• Deep learning excels at classification but faces overfitting and lack of interpretability in high-dimensional, low-sample-size metagenomic datasets.

Purpose of the Study:

• To develop a comprehensive machine learning framework, MetaDR, for accurate and interpretable human disease prediction from microbiome data.
• To integrate diverse data types and deep learning to overcome limitations of existing methods.

Main Methods:

• Developed MetaDR, a machine learning framework integrating various data sources and deep learning.
• Applied MetaDR to metagenomic datasets for human disease prediction.

Main Results:

• MetaDR achieved competitive prediction performance compared to existing methods.
• The framework demonstrated a reduction in running time.
• MetaDR effectively identified informative features with biological relevance.

Conclusions:

• MetaDR offers a robust and efficient approach for microbiome-based disease prediction.
• The framework provides valuable biological insights, addressing the 'black-box' issue of deep learning.
• MetaDR advances the field of metagenomics for clinical applications.