Characterization of vaginal Lactobacillus species as a predictor of fertility among Iranian women with unexplained recurrent miscarriage and fertile women without miscarriage history using machine learning modeling.

Abstract

Lactobacillus spp. are the predominant bacteria of the vaginal tract, the alteration of which has been previously linked to miscarriage. Here, we investigated differences between selected vaginal Lactobacillus species of women with a history of recurrent miscarriages and fertile women without a history of miscarriage in Iran. Vaginal swabs were taken from 29 fertile and 24 infertile women and quantitative real-time PCR (qPCR) assay was used to determine a selection of vaginal Lactobacillus species in both groups. The logistic regression (LR) model, Naive Bayes (NB) model, support vector machine model (SVM), and neural network model (NN) were developed to predict disease outcome by selected variables. LR analysis was used to construct a nomogram indicating predictions of the risk of miscarriage. The most abundant species among the patients were L. rhamnosus, L. ruminis, and L. acidophilus, while L. gasseri, L. vaginalis, L. fermentum, and L. iners were more abundant in healthy subjects. The distribution of L. ruminis, L. iners, and L. rhamnosus was higher in patients, while L. acidophilus, L. gasseri, and L. fermentum were highly distributed among healthy subjects. Higher AUC in predicting the disease outcome was observed for L. gasseri, L. rhamnosus, L. fermentum, and L. plantarum. Our findings provide experimental evidence of vaginal Lactobacillus imbalance in infertile women and a suitable predictor for miscarriage based on the AUC algorithms. Further studies with larger sample size and using high-throughput technologies are needed to boost our understanding of the role of lactobacilli in miscarriage.