MMETHANE: interpretable AI for predicting host status from microbial composition and metabolomics data.

Area of Science:

• Microbiome research
• Metabolomics
• Computational biology

Background:

• Host-microbiome interactions are crucial for health and disease.
• Existing computational tools struggle to link microbiome and metabolomic data to host status.

Purpose of the Study:

• To develop an open-source software package, MMETHANE, for predicting host status from paired microbial and metabolomic data.
• To create an interpretable deep learning model that incorporates biological knowledge.

Main Methods:

• Developed MMETHANE, an open-source deep learning package.
• Incorporated phylogenetic and chemical relationships into the model.
• Trained and validated on six diverse datasets with paired microbial and metabolomic measurements.

Main Results:

• MMETHANE consistently performed on par with or better than existing methods (>80% of datasets).
• The model demonstrated biological interpretability, generating English-language rules.
• Case studies on inflammatory bowel disease revealed significant microbe-metabolite-disease associations.

Conclusions:

• MMETHANE effectively links microbiome and metabolomic data for host status prediction.
• The tool's interpretability aids in understanding complex host-microbiome-metabolome relationships.
• MMETHANE advances the analysis of host-microbiome-metabolome interactions in health and disease.