Adaptive graph learning of microbial phylogeny enables accurate and interpretable microbiome-based host phenotype...

Area of Science:

• Microbiome research
• Computational biology
• Machine learning

Background:

• Human microbiome structure is influenced by microbial phylogeny.
• Current predictive models often overlook evolutionary information, limiting disease classification accuracy.
• Existing deep learning methods may distort phylogenetic relationships by projecting trees into Euclidean spaces.

Purpose of the Study:

• To introduce PhyloGCNE, a framework that models microbiome samples as graphs and integrates phylogeny using edge-aware graph convolution.
• To develop a Phylogenetic Saliency Propagation (PSP) framework for interpretable microbiome analysis.
• To enhance the accuracy and interpretability of microbiome-based host phenotype prediction.

Main Methods:

• PhyloGCNE models microbiome samples as graphs, utilizing edge-aware graph convolution to incorporate phylogenetic information.
• Phylogenetic Saliency Propagation (PSP) is employed for model interpretation, assigning importance scores to microbial taxa.
• The framework was benchmarked on synthetic and eight real-world datasets.

Main Results:

• PhyloGCNE consistently outperformed existing state-of-the-art approaches across diverse datasets.
• The framework demonstrated improved accuracy in predicting host phenotypes, including inflammatory bowel disease and colorectal cancer.
• PhyloGCNE provides interpretable results, identifying specific microbial signatures linked to host phenotypes.

Conclusions:

• PhyloGCNE is an accurate and interpretable phylogeny-aware framework for microbiome analysis.
• Integrating evolutionary information significantly improves microbiome-based host phenotype prediction.
• The PhyloGCNE framework offers a powerful tool for understanding the complex links between the human microbiome and host health.