Faecal microbiome-based machine learning for multi-class disease diagnosis.

Area of Science:

• Microbiome research
• Computational biology
• Disease diagnostics

Background:

• The human gut microbiome plays a role in health and disease.
• Non-invasive diagnostic methods are needed for various conditions.
• Previous single-disease microbiome models face challenges due to overlapping microbial signatures.

Purpose of the Study:

• To develop and validate a machine learning model for multi-disease diagnosis using faecal metagenomic data.
• To assess the model's predictive performance across nine distinct phenotypes.
• To identify microbial species associated with specific diseases.

Main Methods:

• Utilized a faecal metagenomic dataset from 2,320 individuals across nine phenotypes.
• Processed 14.3 terabytes of sequence data covering 325 microbial species.
• Trained and evaluated a machine learning multi-class model, including an independent test set and public datasets.

Main Results:

• The model achieved high predictive accuracy (AUROC 0.90-0.99) in the independent test set.
• Comparable predictions were observed on public datasets (AUROC 0.69-0.91).
• Identified 363 significant associations between microbial species and disease phenotypes (FDR < 0.05).

Conclusions:

• A microbiome-based multi-disease model demonstrates significant potential for non-invasive disease diagnostics.
• The model shows promise for predicting various conditions, including cancer, inflammatory bowel diseases, and cardiovascular disease.
• Further exploration is warranted for clinical application in disease diagnosis and treatment monitoring.