Bayesian dynamic network modelling: an application to metabolic associations in cardiovascular diseases

Area of Science:

• Metabolomics
• Statistical Genetics
• Computational Biology

Background:

• Cardio-metabolic disorders exhibit varying risks across ethnic groups.
• Understanding ethnic differences in metabolite associations over time is crucial for public health.
• The Southall And Brent REvisited (SABRE) study provides a valuable dataset for this research.

Purpose of the Study:

• To develop a novel Bayesian approach for estimating multiple graphical models.
• To analyze temporal patterns of metabolite associations across different ethnic groups (Europeans and South-Asians).
• To identify ethnic-specific differences in metabolite levels and their associations over time.

Main Methods:

• Employing a nodewise regression approach within a Bayesian framework.
• Utilizing the dynamic horseshoe prior to impose sparsity on regression coefficients for graph structure inference.
• Estimating high-dimensional precision matrices of metabolite levels measured at two time points.

Main Results:

• The proposed method allows for the estimation of sparse graphical models reflecting metabolite associations.
• The framework enables the comparison of metabolite network structures across ethnic groups and time points.
• Code is provided for fitting the model using Stan (Hamiltonian Monte Carlo) and a block Gibbs sampling scheme.

Conclusions:

• The novel Bayesian graphical model approach effectively analyzes temporal metabolite associations across ethnicities.
• This method can reveal ethnic-specific insights into cardio-metabolic disorder risk factors.
• The study provides a flexible framework for analyzing complex biological data in cohort studies.