Integrative Statistical Methodology for Metabolomics

• Activity Number: 285
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Francoise Seillier-Moiseiwitsch and Valeriy Korostyshevskiy*+ and Paul Kennedy
• Companies: Georgetown University Medical Center and Georgetown University Medical Center
• Keywords: metabolomics ; alignment ; wavelet ; preprocessing ; test ; multivariate

Abstract

Identifying differentially expressed metabolites is a major challenge in metabolomics. Among the numerous contributors to this challenge are apparatus calibration errors, contamination, as well as misalignment. We have developed comprehensive analytical approaches that deal with preprocessing, alignment and differential analysis. Preprocessing removes the bulk of the background noise. It involves smoothing, selecting regions containing spots and gradient thresholding. Images are aligned using cubic-spline transformations. The alignment is formulated as a quadratic programming problem that is optimized using an interior-point method. We also describe a global approach to feature selection where wavelets are utilized to summarize the aligned images, and statistical tests performed on the wavelet coefficients. Finally, we implement inferential procedures for groups of metabolites.