Computational characterization of bacterial interactions in vaginal microbial communities using the mbtransfer framework

Details

Supervisors

Symul, Laura

Faculty

Faculté des bioingénieurs

Degree label

Master [120] : bioingénieur en sciences agronomiques, à finalité spécialisée

Abstract

enThe vaginal microbiota is central to women's health, with microbial interactions playing a crucial role in maintaining or disrupting vaginal homeostasis. This thesis employs the mbtransfer framework to infer bacterial interactions, leveraging transfer functions and mirror statistics for longitudinal data analysis. Simulation studies were used to evaluate performance under varying ecological architectures, including prey-predator, sparse random, and structured group interactions. Results highlight mbtransfer's efficacy in capturing biologically plausible interactions under conditions of high initial condition heterogeneity and sufficient temporal resolution. However, performance declines with homogeneous initial conditions and inappropriate sampling density or noise, emphasizing sensitivity to study design and the ecological interactions being inferred. Applying the framework to clinical data from the LACTIN-V trial, the inferred networks were able to uncover some literature-validated interactions but also highlighted critical methodological challenges, such as limited baseline diversity and preprocessing sensitivities. Also, the lack of robustness of mirror statistics across multiple runs of the same dataset, and aggressive filtering constrained the ability to detect subtle cross-taxa interactions. These findings underline the importance of optimizing study designs to maximize the diversity and information content of microbial trajectories, and the need for robust computational tools. This work provides insights into improving ecological inference frameworks and advancing microbiome research.