Background: Infertility affects 7-15% of reproductive-age women in the United States. Although in-vitro fertilization (IVF) is widely used, success rates remain limited. The vaginal microbiome, typically dominated by Lactobacillus species, and local inflammatory responses have both been associated with reproductive outcomes, but their combined role in IVF success remains unclear.

Objective: To evaluate the relationship between the vaginal microbiome composition, local inflammation, and IVF outcomes, and to assess their combined predictive value using machine learning.

Methods: We prospectively collected vaginal samples from women undergoing IVF at three time points during the treatment cycle, including participants with unexplained infertility and male factor infertility (MFI). Microbial composition was characterized using 16S rRNA gene sequencing, and concentrations of 20 cytokines were measured in vaginal fluid using multiplex ELISA (Luminex). We developed a supervised machine learning model integrating microbiome and inflammatory data to predict clinical pregnancy and used SHAP analysis to identify key predictive features.

Results: Among 28 participants, 18 achieved clinical pregnancy. Participants who conceived had significantly lower vaginal microbial diversity and lower inflammatory scores. Most individuals with unexplained infertility had Lactobacillus-dominant communities. Among those who conceived, participants with MFI exhibited higher microbial diversity but lower inflammation compared to those with unexplained infertility. Machine learning models achieved the highest predictive accuracy at the peri-retrieval time point, with combined microbiome and inflammatory data improving performance.

Conclusions: Vaginal microbiome composition and host inflammatory responses are key determinants of IVF success. Integrating these features improves prediction of pregnancy outcomes and supports the development of microbiome and immune-based biomarkers to guide personalized fertility care.