Spatial Clustering for Biomarker Discovery in Nuclear Magnetic Resonance-Based Metabolomics Data

• Activity Number: 464
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Xu Han*+ and Genevera I. Allen and Mirjana Maletic-Savatic
• Companies: Baylor College of Medicine and Baylor College of Medicine/Rice University and Baylor College of Medicine
• Keywords: spatial clustering ; biomarker discovery ; pattern recognition ; metabolomics

Abstract

NMR spectroscopy technology produces high-dimensional metabolomic profiles that are mixtures of several latent metabolites often resonating at multiple chemical shifts. To aid in biomarker discovery, the spectrum must be grouped into candidate metabolite regions. We propose to group these variables into spatial clusters, or contiguous sets of variables exhibiting high correlation. The average pair-wise correlation of these contiguous sets is compared to a permutation null distribution obtained by permuting the order of the variables. The optimal threshold used to determine the extent of the spatial clusters is obtained by comparing the average within-cluster correlation to that observed in the original spectrum. We present simulated and real-data examples illustrating the utility of our approach in discovering candidate metabolites from NMR spectroscopy data.