Learning Multiple Gene Regulatory Networks in Type 1 Diabetes through a Fast Bayesian Integrative Method
share
Accurate inference of Gene Regulatory Networks (GRNs) is pivotal to gaining a systematic understanding of the molecular mechanism and identifying potential therapeutic targets for diseases. Many methods such as Gaussian Graphical Models have been proposed for explore conditional independence relationships in the networks. However, the application of such methods are limited due to the dynamic involvement of gene regulation networks when data are collected under distinct conditions. A motivating example is that of modeling gene expressions in The Environmental Determinants of Diabetes in the Young (TEDDY) study, where the expression data from one participant can be collected on multiple time points and attributed to different comparison groups, hence, gene regulation networks under distinct conditions tend to be dependent with each other. In this paper, we proposed a innovative method for jointly estimating multiple dependent Gaussian graphical models. First, we adopt a ψ-learning algorithm to transfer the expression data to the pair-wise scores. Then, we propose a Bayesian clustering and meta-analysis method to identify the possible edge changes and integrate the pair-wise scores. Finally, we apply a multiple hypothesis test to determine the gene network structures under each condition. The result identifies several important genes related to the development of Type 1 Diabetes and reveals the evolvement of gene networks throughout time.
READ FULL TEXT
