Title: Modelling ethnic differences in metabolic associations via dynamic Bayesian nodewise regression
Authors: Maria De Iorio - UCL (United Kingdom) [presenting]
Abstract: A novel approach is proposed to the estimation of multiple Gaussian Graphical Models to analyse dynamic evolving patterns of association among a set of metabolites over different groups of patients. The motivating application is the Southall And Brent REvisited study, a tri-ethnic cohort study conducted in the UK. We are interested in identifying potential ethnic differences in metabolite levels and associations, with the aim of gaining a better understanding of different risk of cardio-metabolic disorders across ethnicities. We model the inverse-covariance structure of a set of metabolites measured over different time points and for three ethnic groups. We adopt a Bayesian adaptation of the Nodewise Regression technique to infer the structure of the graphs. We assume a global-local shrinkage prior over the regression parameters to impose a sparse structure on the graph, and we extend the prior to allow borrowing of information across different groups. Finally, we extend the model to a dynamic framework to impose a time dependence and estimate multiple graphs over time. Posterior inference is performed through Markov Chain MonteCarlo methods. Specifically, we use the software Stan, which employs Hamiltonian Monte Carlo. The proposed approach is able to capture a wide range of graph topologies and identify diverse/common structures across multiple graphs, corresponding to different ethnicities and allows us to estimate time trends for each metabolite connection.