Modeling the human vaginal microbiome and its protection against pathogens using the replicator framework for invasion

Abstract

The human vaginal microbiota plays a central role in protecting against urogenital infections, including bacterial vaginosis, yeast infections, HIV, and urinary tract infections. However, the ecological mechanisms connecting microbial community structure to clinical indicators such as Nugent scores remain poorly understood. Although machine-learning approaches can accurately predict bacterial vaginosis (BV) from microbiota profiles, they provide limited biological insight. Here, we introduce a mechanistic framework that links vaginal microbiota composition to Nugent score and identifies key ecological interactions underlying health and disease-associated community states. We analyzed microbiota data from an already published North American women cohort, aggregating taxa at the phylum level to improve stability and interpretability. Using a replicator model, we quantified both the direct contributions of individual phyla and nonlinear effects arising from pairwise interactions, capturing transitions among healthy, intermediate, and BV-positive states. The model predicted BV-positive status with 92% accuracy for the 394 women in the study, on par with machine-learning benchmarks. More importantly, it provides a clear ecological interpretation of BV-associated community change. Beyond the BV setting, the model itself illustrates a general proof-of-concept for microbiota-invader links via the replicator formalism.