Multi-cohort ensemble learning framework for vaginal microbiome-based endometrial cancer detection.

Area of Science:

• Gynecologic Oncology
• Microbiome Research
• Computational Biology

Background:

• Endometrial cancer is a common gynecological malignancy lacking early detection strategies.
• The vaginal microbiome's diagnostic potential for endometrial cancer is suggested but requires further validation.

Purpose of the Study:

• To identify microbial signatures associated with endometrial cancer.
• To develop a predictive machine learning model for early endometrial cancer detection using vaginal microbiome data.

Main Methods:

• Systematic review and analysis of vaginal 16S rRNA sequencing data from five independent cohorts (n=265).
• Comparison of microbial diversity and composition between endometrial cancer patients and controls.
• Development and validation of a machine learning classifier.

Main Results:

• Significantly higher microbial diversity in endometrial cancer samples.
• Reproducible enrichment of *Peptoniphilus* in endometrial cancer samples across cohorts.
• An ensemble classifier achieved high accuracy (AUC 0.93), sensitivity (1.0), and negative predictive value (1.0) in identifying endometrial cancer.

Conclusions:

• Vaginal microbiome profiling is a potential minimally invasive approach for early endometrial cancer detection.
• Machine learning models can effectively utilize microbial signatures for cancer diagnosis.