MITRE: inferring features from microbiota time-series data linked to host status.

Area of Science:

• Microbiome research
• Computational biology
• Machine learning applications in health

Background:

• Longitudinal studies are essential for understanding the microbiome's role in human disease.
• Current machine learning methods for microbiome analysis often lack interpretability.

Purpose of the Study:

• To introduce MITRE (Microbiome Interpretable Temporal Rule Engine), a novel supervised machine learning method.
• To enable the discovery of biologically interpretable relationships between microbiome dynamics and host status.

Main Methods:

• Developed MITRE, a supervised machine learning approach for microbiome time-series analysis.
• Inferred human-interpretable rules linking microbial clade abundance changes over time to host disease status (presence/absence).
• Validated MITRE using semi-synthetic data and five real-world microbiome datasets.

Main Results:

• MITRE demonstrated performance on par with or superior to conventional, less interpretable machine learning methods.
• The method successfully identified interpretable rules from complex microbiome time-series data.
• Validated findings across diverse datasets, confirming robustness and generalizability.

Conclusions:

• MITRE provides a powerful new tool for microbiome time-series analysis.
• Enables the discovery of biologically meaningful and interpretable links between the microbiome and human health.
• Facilitates deeper understanding of host-microbiome interactions in disease contexts.