Predicting Bacterial Vaginosis Development using Artificial Neural Networks
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Bacterial vaginosis (BV) is a dysbiosis of the vaginal microbiome, characterized by the depletion of protective Lactobacillus spp. and overgrowth of anaerobes. Artificial neural network (ANN) modeling of vaginal microbial communities offers an opportunity for early detection of incident BV (iBV). 16S rRNA gene sequencing and quantitative PCR was performed on longitudinal vaginal specimens collected from participants within 14 days of iBV or from healthy participants to calculate the inferred absolute abundance (IAA) of vaginal bacterial taxa. ANNs were trained using the IAA of vaginal taxa from 420 vaginal specimens to classify individual vaginal specimens as either pre-iBV (collected before iBV onset) or Healthy. Feature importance was assessed to understand how specific vaginal micro-organisms contributed to model predictions. ANN modeling accurately classified >97% of specimens as either pre-iBV or Healthy (sensitivity >96%, specificity >98%) using IAA of 20 vaginal taxa. Model prediction accuracy was maintained when training models using only a few key vaginal taxa. Models trained using only the top five most important features achieved an accuracy of >97%, sensitivity >92%, and specificity >99%. Model predictive accuracy was further improved by training models on specimens from white and black participants separately; using only three feature models achieved an accuracy >96%, sensitivity >91%, and specificity >91%. Feature analysis found that Lactobacillus species L. gasseri and L. jensenii differed in how they contributed to model predictions in models trained with data stratified by race. A total of 420 vaginal specimens were analyzed, providing a robust dataset for model training and validation.
Importance
Bacterial vaginosis (BV) is the most common vaginal infection and is associated with numerous comorbidities. BV is associated with infertility, preterm birth, pelvic inflammatory disease, and increased risk of HIV/STI acquisition. BV is difficult to detect prior to onset, and infection commonly recurs after treatment. Our model allows for the accurate early detection of iBV by surveying the vaginal microbiome, potentially serving as a valuable tool to determine which patients are at risk of developing iBV. Early detection of iBV could lead to wider adoption of clinical interventions useful in the prevention of iBV such as live biotherapeutics, prophylactic antibiotics, and/or behavioral modifications. Our findings indicate that few microbial targets are required for accurate predictions, facilitating cost and time effective clinical testing. Similarly, our study highlights the value of developing models personalized to specific patient populations, improving accuracy while reducing the number of taxa required for accurate predictions.
