A0417
Title: Joint estimation of multiple Gaussian graphical models via multiple hypothesis tests
Authors: Faming Liang - University of Florida (United States) [presenting]
Abstract: Gaussian Graphical Models have been widely used to explore conditional independence relationships for a large set of random variables. We propose a new method for jointly estimating multiple Gaussian graphical models with observations belonging to distinct classes, which works under the framework of multiple hypothesis testing and includes an meta-analysis procedure to explicitly integrate the data information across distinct classes. The proposed method has some significant advantages over the existing ones. First, it provides a more explicit and sufficient way to integrate the data information across multiple classes. However, the existing methods often integrate the data information through prior distributions or penalty function, and this is often less sufficient. Second, it can provide an uncertainty measure for the edges detected in the multiple graphical models and the difference of edges detected in the graphical models under any two distinct conditions, while the existing methods only produce a point estimate or are feasible for very small size problems. We illustrate the performance of the proposed method using simulated and real data examples. The numerical results indicate the superiority of the proposed method over the existing ones.
