DeepBiome

DeepBiome employs a phylogeny-informed deep neural network to analyze microbiome abundance (bacterial count) data and predict host phenotypes for human health research.

Key Features:

• Phylogeny-Informed Neural Network Architecture: Incorporates evolutionary relationships between bacteria via a phylogenetic tree into the neural network design to improve phenotype prediction from microbiome counts.
• Handling Sparse Data and Taxonomic Hierarchies: Operates on sparse bacterial count tables and is not restricted to predefined taxonomic levels, enabling analysis across taxonomic ranks from species to phylum.
• Reduced Need for Extensive Training Data: Utilizes phylogenetic guidance to mitigate the requirement for very large training datasets and to reduce the risk of overfitting in small to moderate sample sizes.
• Regression and Classification Support: Applicable to both regression and classification tasks for microbiome-based phenotype modeling.
• Visualization and Interpretation: Provides visualization of pathways linking microbiome counts to diseases to support interpretation of microbiome-phenotype associations.

Scientific Applications:

• Phenotype Prediction: Predicts host phenotypes from microbiome abundance data using phylogeny-informed models.
• Microbiome-Phenotype Association Discovery: Uncovers associations within the microbiome-phenotype network and provides ecological and evolutionary insights into host-microbe interactions.
• Biomedical Regression and Classification Studies: Supports regression and classification analyses in human health research using microbiome data.
• Validation on Simulations and Real Data: Has been evaluated on simulation studies and real-life data analyses demonstrating accuracy and efficiency.

Methodology:

Phylogeny-informed neural network architecture implemented in Python using Keras and TensorFlow that integrates phylogenetic tree information with microbiome abundance (bacterial count) tables.