SMOTE-augmented machine learning model predicts recurrent and metastatic breast cancer from microbiome analysis

Area of Science:

• Microbiome Research
• Oncology
• Bioinformatics

Background:

• Recurrence and metastasis of breast cancer (RMBC) significantly impact patient survival.
• Reliable biomarkers are crucial for early prediction of RMBC.
• Blood microbiome composition is increasingly recognized for its potential health implications.

Purpose of the Study:

• To evaluate blood microbiome profiles as predictive biomarkers for RMBC.
• To apply machine learning techniques for early RMBC prediction.
• To identify specific microbial signatures associated with RMBC.

Main Methods:

• Retrospective analysis of 288 participants (96 breast cancer patients, 192 controls).
• 16S rRNA sequencing for blood microbiome analysis.
• Machine learning models (including random forest) trained and validated using cross-validation techniques.
• SMOTE applied for class imbalance; performance evaluated using AUROC, recall, precision, F1-score, and MCC.

Main Results:

• Alpha diversity was significantly lower in disease-free survival (DFS) and RMBC groups compared to controls (p < 0.05).
• Distinct clustering observed in beta diversity analysis.
• Random forest model achieved high performance: AUROC 0.94, recall 0.81, F1-score 0.83, MCC 0.88.
• Key genera predicting RMBC included Enterobacter, Bacteroides, Klebsiella, and Bifidobacterium.

Conclusions:

• Blood microbiome profiling demonstrates potential as a non-invasive biomarker for RMBC.
• Machine learning models can effectively distinguish RMBC based on blood microbiome data.
• Further validation studies are warranted to confirm these findings.