Séminaire
Organisme intervenant (ou équipe pour les séminaires internes)
MIA PS
Nom intervenant
Alexandre Chaussard
Résumé
The gut microbiome is a complex ecosystem with a strong influence on its host's health. High-throughput sequencing pipeline allow to explore its composition by providing high-dimensional count profiles that characterize the abundance and diversity of microbial taxa. However, analyzing microbiome profiles remains challenging due to compositional structures, sparsity, pronounced overdispersion, measurement noise, and external perturbations.
In this presentation, we focus on a probabilistic modeling of temporal count data based on Poisson Log-Normal models. Building on structured independent component analysis (ICA) with latent autoregressive dynamics, we aim to learn a disentangled, low-dimensional representation that supports downstream analysis while offering an interpretable view of microbial co-variations. To capture time-varying dynamics, we couple this representation with a switching linear dynamical system, enabling regime inference and modeling of perturbation effects along time. We illustrate the approach on a gnotobiotic mice experiments, and discuss identifiability results that underpin the interpretability of the recovered components and the ICA mixing function.
Lieu
Amphi C2 (peut varier voir mail d'annonce)
Date du jour
Date de fin du Workshop