In recent years, novel methods for graphical model inference have been widely applied to infer biological networks for high throughput data. When studying complex diseases, these network-based inferential approaches can be crucial in evaluating patterns of variable association and determining the cellular level changes influencing disease susceptibility, progression, and variation across heterogeneous subjects. This thesis is focused on developing flexible joint graph methodology for estimating network structures of multiple sample groups. These approaches employ Gaussian graphical models, Markov random field priors, continuous shrinkage priors, and Dirichlet process priors to infer graph structures for each sample group, while sharing information between related graphs without assuming any similarity. These methods are illustrated through simulation studies and applications to case studies of Chronic Obstructive Pulmonary Disease (COPD).