Yi Li

A New Framework for Integrative Analyses

Understanding how genes interplay  with each other and how their regulations are associated with high-dimensional genomic markers may uncover the underlying mechanism of disease progression processes.  Gaussian graphical models have been used in simultaneously learning the response network and the associations between the response. These models assume a homogeneous population and ignore the heterogeneity between individual-level networks. Limited work has been done to detect associations between response network structures and high-dimensional predictors. In this talk, we propose a conditional graphical model with functional precision parameters. We propose a Fisher scoring matching approach for variable selection and network recovery.  We show that the proposed method can consistently select important predictors and recover the response network structure. The proposed method is computationally inexpensive and can be directly applied to analyzing ``omic" scaled networks and DNA data, such as the cancer genome atlas (TCGA) data, to study cancer-triggering biological pathways.