TaxoNN

TaxoNN predicts disease status from microbiome data by stratifying Operational Taxonomic Units (OTUs) at the phylum level, training phylum-specific Convolutional Neural Networks (CNNs), and integrating their outputs with ensemble learning to improve predictive accuracy.

Key Features:

• Stratified OTU grouping: Groups OTUs by phylum within the microbial taxonomy to account for hierarchical correlations among taxa.
• Phylum-specific CNNs: Trains Convolutional Neural Networks independently within each phylum cluster to capture intra-phylum OTU patterns.
• Ensemble feature integration: Concatenates features from each phylum-specific CNN and integrates them via ensemble learning for final prediction.
• Performance validation: Evaluated on simulated and human microbiome datasets (168 OTUs, 200 cases and controls), reporting mean AUC improvements over Random Forest of approximately 5%, 3%, and 7% across tested datasets, including cirrhosis and type 2 diabetes.

Scientific Applications:

• Precision medicine: Infers disease-associated microbiome signatures to support targeted interventions and risk stratification.
• Early diagnosis and personalized treatment: Predicts disease risk from microbiome composition to inform early diagnosis and individualized therapy decisions.
• Disease-specific microbiome studies: Applied to cirrhosis and type 2 diabetes datasets to evaluate microbiome-based predictive biomarkers.

Methodology:

OTUs are stratified into phylum-level clusters; Convolutional Neural Networks are trained separately within each cluster; features from each CNN are concatenated and combined using ensemble learning for final prediction.