A Goldilocks Principle for the Gut Microbiome: Taxonomic Resolution Matters for Microbiome-Based Classification of...

Area of Science:

• Microbiome research
• Cancer diagnostics
• Machine learning applications

Background:

• Colorectal cancer (CRC) is a leading cause of cancer deaths in the U.S.
• Early detection and treatment are crucial for CRC prevention
• The gut microbiome offers potential for noninvasive CRC detection

Purpose of the Study:

• To determine the optimal taxonomic resolution for microbiome-based CRC classification
• To evaluate the impact of different taxonomic levels on machine learning model performance

Main Methods:

• Utilized a machine learning framework to analyze microbiome data
• Compared classification performance across taxonomic resolutions from phylum to ASV
• Quantified model performance using area under the receiver operating characteristic curve (AUROC)

Main Results:

• Model performance increased with resolution up to the family level
• Family, genus, and operational taxonomic unit (OTU) levels showed similar, optimal performance (AUROC ~0.69)
• Amplicon sequence variant (ASV) level resolution significantly decreased performance (AUROC ~0.676)

Conclusions:

• Mid-range taxonomic resolution (family, genus, OTU) is optimal for CRC prediction from microbiome data
• Excessively fine taxonomic resolution (ASV) can hinder accurate sample classification
• Appropriate taxonomic resolution is critical for effective microbiome-based cancer diagnostics