Alterations in the cervicovaginal microbiome and their diagnostic potential in cervical cancer

The cervicovaginal microbiome is increasingly implicated in cervical carcinogenesis, yet practical diagnostic tools leveraging microbial signatures remain underdeveloped. This study aimed to characterize the cervicovaginal microbial ecology in cervical disease and evaluate the feasibility of a microbiome-based machine learning model for clinical classification. We conducted 16 S rRNA gene sequencing on vaginal swabs from 163 patients (cervical cancer and precancerous lesions) and 70 healthy controls. Microbial diversity, taxonomic composition, and functional pathways were compared between groups. A random forest classifier was constructed using selected microbial biomarkers, with performance validated across training, test, and independent cohorts. Patients exhibited significantly increased alpha diversity (Shannon, Chao1, ACE) and distinct beta diversity (PCoA/NMDS, p < 0.05) versus controls. Taxonomic analysis revealed enrichment of Prevotella , Streptococcus , and Pseudomonas , alongside depletion of Fannyhessea and Sneathia . Functional predictions showed upregulation of amino acid and energy metabolism, and downregulation of transport/signaling pathways in patients. The nine-biomarker random forest model achieved robust AUCs of 0.891 (training), 0.896 (test), and 0.825 (independent validation). Cervical disease is associated with a distinct and dysbiotic cervicovaginal microbiome, with compositional and functional reprogramming. The microbial biomarker model demonstrates strong diagnostic accuracy, supporting its potential as a complementary, non-invasive screening tool for cervical cancer and precursors.