Elucidating the physiological and pathological characteristics of preterm birth is critical to the development of effective prediction and intervention strategies. Here, we present a nested case control study on 51 spontaneous preterm birth (sPTB) and 70 women who delivered at term. Systems-level analysis revealed bacterial compositions and functional differences between two distinct pregnancy outcomes. As expected, the vaginal environment with preterm birth was dominated by a more complex bacteria interaction network, and these interactions frequently exhibited opposite effects on different pregnancy outcomes. Furthermore, key species and combinations were identified to construct preterm birth prediction model. The final model contained three features generated from six bacteria (A. christensenii and L. crispatus; B. breve and S. anginosus; P. timonensis and L. fermentum) (AUROC = 0.942 for independent test). The AUROC value was 0.965 when white blood cell (WBC) and neutrophil (NEU) data were included. Collectively, our study provided a comprehensive characterization of the vaginal microbiome and revealed the potential utilization of bacteria in predicting the risk of preterm birth.