Symul, LauraRahali Masbouri, MariamMariamRahali Masbouri2025-07-012025-07-012025-06-1020252025-06-10https://hdl.handle.net/2078.2/43286Antibiotic treatments like metronidazole (MDZ) induce significant changes in the vaginal microbiota, influencing treatment efficacy and bacterial vaginosis (BV) recurrence. This thesis applies the mbtransfer framework, a causal inference method using transfer functions and mirror statistics, to identify taxa affected by MDZ and characterize recovery dynamics. Simulation studies demonstrated that mbtransfer reliably detects intervention effects under low-to-moderate noise, particularly with large effect sizes and numerous affected taxa. Power analyses indicate that sample sizes of at least 40 participants are necessary for robust inference in noisy settings. Applied to LACTIN-V clinical data, mbtransfer identified significant post-treatment declines in several non-Lactobacillus taxa, including Fannyhessea vaginae, Prevotella spp., and BVAB1. Lactobacillus iners appeared largely resilient to MDZ, while Gardnerella vaginalis showed moderate susceptibility, suggesting the presence of a susceptible clade despite expectations of resistance. These results support the role of species-specific susceptibility and resilience in BV recurrence. This study highlights the benefit of combining causal modeling with ecological insights to understand microbiota responses to antibiotics. Future work should focus on benchmarking mbtransfer, modeling delayed antibiotic effects, refining ecological interactions, incorporating resistance dynamics, and integrating multimodal data for comprehensive host-microbiota analysis.vaginal microbiotabacterial vaginosismetronidazoletransfer functionsmirror statisticscausal inferencetext::thesis::master thesis