MITRE

MITRE infers interpretable temporal rules that link changes in microbial clade abundances in microbiome time-series to binary host-status outcomes such as disease presence or absence.

Key Features:

• Interpretable Rules: Generates short lists of human-interpretable rules that relate temporal changes in microbial clade abundances within defined time windows to host health conditions.
• Temporal Analysis: Analyzes longitudinal microbiome time-series to identify dynamic interactions between the microbiome and binary host status.
• Validation on Diverse Data Sets: Validated on semi-synthetic data and five real-world datasets, demonstrating performance comparable to or exceeding conventional machine learning approaches while maintaining interpretability.

Scientific Applications:

• Microbiome longitudinal studies: Dissects temporal dynamics of microbial communities to elucidate relationships between microbiome changes and human health.
• Translational discovery: Supports identification of potential causal links that may inform therapeutic strategies or diagnostic tool development.

Methodology:

Applies supervised machine learning to microbiome time-series to detect temporal patterns in microbial clade abundances and generate interpretable rules linking those patterns to binary host-status outcomes (e.g., disease presence/absence).