A Kernel Machine Approach for Metabolomics

• Activity Number: 400
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Xiang Zhan*+ and Debashis Ghosh and Andrew Patterson
• Companies: Penn State and Penn State and Penn State
• Keywords: distance-based kernel ; metabolomics ; missing value ; score test ; stratified kernel

Abstract

A common phenomenon in metabolomics measurements is that the data matrix frequently contains missing value. The primary methods of dealing with missingness so far have been methods that attempts to impute the missing values. However, one may argue that this fails to incorporate the missingness as a fundamental feature of the data. An alternative way is to treat a missing value as absence of a metabolite. Hence there are two types of information that are available in metabolomics data: presence/absence of a metabolite and a quantitative value of the metabolite if it is present. These two layers of information pose challenges to application of traditional statistical approaches in differential expression analysis. We propose a kernel machine-based score test for the metabolomics differential expression analysis. In order to capture the special pattern of metabolomics data, two new kernel machines are designed. One is the distance-based kernel and the other is the stratified kernel. Those methods are illustrated with application to simulated data as well as data from a liver cancer metabolomics study. They are shown to have better performance than some existing alternatives.