Novel Methodologies for Gene Network Interaction Analysis

• Activity Number: 534
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Sang-Yun Oh*+ and Bala Rajaratnam
• Companies: Stanford University and Stanford University
• Keywords: high dimensional inference ; network models ; covariance ; multivariate analysis

Abstract

The availability of high-throughput data in biomedical applications has created an urgent need for methodology for analyzing high-dimensional data. Complex relationships between variables (genes/proteins) in high dimensional data are often understood in terms of networks/pathways. The covariance parameter (or its inverse) is a natural parameter of interest when trying to understand such complex relationships between many variables. Here network models can be very useful in capturing the essence of the main interactions between many variables. In this paper we study a new method to estimate the inverse covariance matrix in a sparse manner that is suitable for genomic and biomedical applications, and thus estimate the underlying biological networks which are present in such data sets. We compare our method to others in the literature to assess its effectiveness in high dimensional problems