B0506
Title: Unravelling complex diet-gut microbiome-host health interaction by mixture of experts models
Authors: Xiangnan Xu - Humboldt University of Berlin (Germany) [presenting]
Sonja Greven - Humboldt University of Berlin (Germany)
Muller Samuel - Macquarie University (Australia)
Abstract: The gut microbiome is crucial for human health, influenced by various factors, particularly diet. However, the relationships between diet, the gut microbiome and host health are complex and heterogeneous. Individuals with different diets can provide distinct sources of energy, which impact the association between microbiome and host health. To unravel these relationships, two models are proposed: a nutrition-ecotype graphical mixture of experts (NEGMoE) and the nutrition-ecotype mixture of experts (NEMoE) models. NEGMoE focuses on microbial co-abundance networks and incorporates diet-specific cohort variability via a mixture of expert (MoE) models. It uses a graphical lasso penalty to identify nutritional subcohorts and determine the distinct microbial relationship correlations within each subcohort. Meanwhile, NEMoE also utilizes the MoE approach to deal with the differential relationship between the microbiome and health outcomes. By optimizing these models, diet-specific subcohorts with differential microbial relationships and microbial disease signatures are identified. NEGMoE and NEMoE are applied to both simulated and real-world microbiome datasets. Simulation studies showed that NEGMoE and NEMoE could robustly identify subcohorts with different correlation structures and relationships with response variables. In the real-world data, NEGMoE and NEMoE identified biologically meaningful subcohort with diet-specific correlation structure and microbial disease signature.
