Related Experiment Video
Updated: Sep 28, 2026
Modeling Healthy and Dysbiotic Vaginal Microenvironments in a Human Vagina-on-a-Chip
Published on: February 16, 2024
Multi-omics integration of the vaginal microbiome and inflammatory proteome for non-invasive prediction of
Yangkun Feng1,2,3, Yu Zhang1,2,3, Dandan Chen1,2,3
1Department of Reproductive Medicine Center, Wuxi Maternity and Child Health Care Hospital, Wuxi, China.
Background:
Infertility remains a global health challenge that often necessitates invasive and costly diagnostic procedures. While the vaginal microenvironment is known to influence reproductive health, its potential as a source of non-invasive biomarkers for infertility remains insufficiently explored. This study integrated vaginal microbiome and inflammatory proteomic profiling using machine learning to develop and validate an objective, non-invasive diagnostic model for identifying women at high risk of infertility.
Methods:
A total of 205 women (76 infertility patients and 129 healthy controls) were enrolled between October 2025 and February 2026. Vaginal swabs were collected for 16S rRNA sequencing and Olink Target 96 Inflammation panel analysis. LASSO, Boruta, and RFE were applied for feature selection, and eight machine learning algorithms were established to the training cohort (n = 143). Model performance was validated in the test cohort (n = 62) using AUC and DeLong tests. SHAP analysis was utilized for biological interpretability.
Results:
4 Olink features and 14 microbial features were retained for analysis. The microbiome-based SVM model demonstrated the highest predictive performance with an AUC of 0.815 (95% CI: 0.710-0.920), significantly outperforming the proteomics-based XGBoost model (AUC = 0.629, 95% CI: 0.489-0.768). Multi-omics based SVM model yielded an AUC of 0.799 (95% CI: 0.691-0.993). SHAP analysis identified the genera Thomasclavelia, unclassified Ruminococcaceae, and Megamonas as the top 3 contributing features in the predictive model. Furthermore, all Olink biomarkers retained in the prediction model, including MMP-1, MMP-10, CCL20, and CXCL5, were identified as contributing risk factors associated with infertility.
Conclusion:
The vaginal microbiome is a potent non-invasive biomarker for infertility, offering superior diagnostic accuracy compared to host inflammatory proteins. This machine learning-based multi-omics approach provides a promising tool for early risk stratification and personalized reproductive management.
