G2S

G2S predicts the taxonomic structure of the human stool microbiome from oral microbiome data using deep learning for metagenomic inference.

Key Features:

• Deep convolutional neural network (CNN): Implements a deep CNN architecture trained on comprehensive datasets from the Human Microbiome Project.
• Training data source: Model training uses Human Microbiome Project datasets to learn relationships between oral and stool microbiomes.
• Taxonomic resolution: Produces inferred stool microbiome composition at the family taxonomic level.
• Oral-to-stool prediction: Infers stool microbiome structure directly from an individual's oral microbiome profiles.
• Validation on paired samples: Model performance was validated using characterized pairs of oral and fecal samples.
• Application to ancient samples: Has been applied to dental calculus (ancient microbiome) to reconstruct intestinal components of medieval populations.

Scientific Applications:

• Modern sample analysis: Predicts eubiotic or dysbiotic states of the gut microbiome from oral samples when fecal sampling is unavailable.
• Paleomicrobiology and ancient health: Reconstructs gut microbiome configurations from dental calculus to investigate intestinal components of historical populations.

Methodology:

Uses a deep convolutional neural network trained on Human Microbiome Project datasets to infer stool microbiome composition at the family level from oral microbiome data, validated on paired oral and fecal samples and applied to dental calculus datasets.