Multi-omics machine learning to study host-microbiome interactions in early-onset colorectal cancer

Area of Science:

• Oncology
• Microbiome Research
• Metabolomics
• Machine Learning

Background:

• Early-onset colorectal cancer (eoCRC) incidence is increasing, with unclear pathogenesis.
• Host-microbiome interactions are implicated in CRC development.
• Understanding unique associations in eoCRC is crucial for targeted interventions.

Purpose of the Study:

• To investigate distinct host-microbiome associations in eoCRC versus average-onset CRC (aoCRC) using machine learning.
• To identify potential biomarkers for eoCRC by integrating microbiome and metabolome data.
• To explore therapeutic intervention opportunities based on identified host-microbiome correlations.

Main Methods:

• Paired analysis of tumor tissue microbiome (16S rRNA sequencing) and plasma metabolome in CRC patients (n=64).
• Categorization into eoCRC (age ≤ 50) and aoCRC (age ≥ 60).
• Application of DIABLO machine learning for multi-omics data integration and classifier development; differential association network analysis.

Main Results:

• Multi-omic dimension reduction revealed distinct clustering patterns between eoCRC and aoCRC.
• A metabolomics classifier achieved high diagnostic performance (AUC=0.98), outperforming the microbiome classifier (AUC=0.61).
• Specific microbial-metabolite correlations were identified, including glycerol and pseudouridine with Parasutterella/Ruminococcaceae, and cholesterol/xylitol with Erysipelatoclostridium/Eubacterium/Acidovorax.

Conclusions:

• Multi-omics analysis effectively reveals host-microbiome correlations in eoCRC.
• A metabolomics classifier demonstrates significant potential as a biomarker for eoCRC.
• Distinct host-microbiome correlations, particularly involving the urea cycle in eoCRC, may present novel therapeutic targets.