Understanding the complex interactions between the microbiome, the host, and external factors is essential for uncovering how dysbiosis mediates the effects of interventions or environmental exposures on health outcomes. However, analyzing microbiome count data, particularly with high zero inflation, presents significant challenges. Zero-inflated models often struggle to accurately estimate the proportion of zeros due to the reduced number of non-zero counts, limiting their effectiveness. To address this, we developed a Bayesian framework using a zero-inflated negative binomial model for microbiome counts. We introduced spike-and-slab priors to induce sparsity in mediation effects and incorporated informative priors based on non-zero counts to improve dispersion estimation and control the source of zeros. Additionally, we separated natural indirect effects (NIE) into two distinct pathways—mediator abundance and prevalence—creating a dual mediation model that captures different mediation mechanisms for rare and abundant species. This dual approach enhances testing power and improves biological interpretation. Simulations show superior performance compared to existing methods, and its application to human microbiome studies, such as nutrition intake and metabolic syndrome, highlights its effectiveness in identifying key microbial mediators, providing valuable insights into disease physiology and health.
Zoom link: https://uoc-gr.zoom.us/j/88659969718?pwd=g6bjYPDCuUQo1bzVxjjbgQL4xFN1f3.1
