Network-Based Integrative Analysis and Marker Selection with Ultrahigh Dimensional Data

• Activity Number: 663
• Type: Invited
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Shuangge Ma*+
• Companies: Yale University
• Keywords: Ultrahigh dimensional data ; integrative analysis ; marker selection ; network analysis

Abstract

With the fast development of data collection techniques, ultrahigh dimensional data are now commonly encountered in finance, engineering, biomedicine and other scientific areas. Marker selection is usually needed in the analysis of such data. Our study shows that marker selection results from the analysis of a single dataset can be unreliable because of the high dimensional nature of variables. Integrative analysis provides an effective way to pool and analyze multiple heterogeneous studies, and can outperform single-dataset analysis. We have developed a network based integrative analysis approach. In network analysis, a node corresponds to a variable; and two nodes are connected if the corresponding variables have similar functionalities or correlated measurements. The proposed approach advances from the existing integrative analysis approaches by properly accounting for the network connection measure. Numerical studies, including simulation and analysis of biomedical data, show that the proposed approach outperforms alternatives in terms of marker selection accuracy and prediction performance.