Penalized Estimation for Sparse Gaussian Concentration Graphs with Applications to Inference of Genetic Regulatory Networks

• Activity Number: 87
• Type: Invited
• Date/Time: Monday, August 8, 2005 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Hongzhe Li*+ and Jiang Gui
• Companies: University of California, Davis and University of California, Davis
• Keywords: gene regulatory pathways ; threshold gradient descent ; penalized likelihood ; covariance selection ; gene expression ; microarray

Abstract

Large-scale microarray gene expression data provide the possibility of constructing the gene regulatory networks or pathways. It has been noted that Gaussian graphical models provide less-restrictive alternatives to Bayesian networks in modeling such regulatory networks. However, such models are computationally more demanding, especially in the setting of high-dimension and low sample size. In this talk, we introduce a threshold gradient descent procedure for penalized estimation of sparse concentration matrix in the setting of Gaussian graphical models. We also demonstrate its application to identifying gene regulatory networks. Such procedure easily can incorporate biological knowledge about network structure. We present our results on inference, a gene network for isoprenoid biosynthesis in Arabidopsis thaliana. A comparison with the SINful approach of Drton and Perlman and an extension to time course microarray gene expression data also is presented and discussed.