Leveraging machine learning techniques to investigate the pathogenesis of incident bacterial vaginosis

• Author(s): Elnaggar, J. ; Jacobs, C. ; Ardizzone, C. ; Aaron, K. ; Eastlund, I. ; Graves, K. ; Luo, M. ; Tamhane, A. ; Long, D. ; Laniewski, P. ; Herbst-Kralovetz, M. ; Quayle, A. ; Cerca, Nuno ; Muzny, C. ; Taylor, C.
• Date: 2024
• Persistent ID: https://hdl.handle.net/1822/94179
• Origin: RepositóriUM - Universidade do Minho
• Document Type: Other
• Language: English
• Contributor(s): Universidade do Minho

Description

Objective We investigated longitudinal changes in the vaginal microbiota prior to, during, and immediately following the onset of incident bacterial vaginosis (iBV) to investigate its pathogenesis. We used supervised machine learning and artificial neural networks (ANNs) to identify vaginal micro-organisms predictive of future iBV development. Study Design In this prospective cohort study, we performed 16S rRNA gene sequencing on twice-daily longitudinal vaginal specimens from cisgender women who developed iBV (Nugent score 7-10 on 4 consecutive specimens) and matched healthy controls maintaining a normal vaginal microbiota (Nugent score 0-3) for the majority of the study. ANNs were used to associate changes in the inferred absolute abundance (IAA) of select vaginal bacteria to investigate iBV pathogenesis. Results We analyzed 212 vaginal specimens from 8 participants with iBV, including up to 14 days pre-iBV (n=127), during iBV (n=57), and 5 days post-iBV (n=15), alongside 283 specimens from 8 healthy controls. The ANN suggested that Gardnerella spp., L. crispatus, and L. iners were the most predictive micro-organisms of whether a participant develops iBV or maintains a normal vaginal microbiota. Investigating these micro-organisms individually, we observed a significant decrease in the IAA of L. crispatus pre-iBV (p<0.001), an increase in Gardnerella spp. prior to and during iBV (p<0.001), and an increase in L. iners post-iBV (p<0.001). Conclusion This study found that the ratios of Gardnerella spp. (iBV-associated), L. crispatus (protective), and L. iners (recovery) were most important for predicting optimal versus BV microbiota. These findings may guide future BV diagnostic tests and BV pathogenesis research.