Statistical Engineering Division Seminar

Dr. Katharine M. Mullen
Department of Physics and Astronomy
Vrije Universiteit Amsterdam
Monday, May 19, 2008, 10:45-11:45 AM
Building 222, Room A330

Abstract

Mass spectrometry data typically represent the superposition of the contribution of many distinct chemical compounds, most often resolved with respect to both time and mass-to-charge ratio. Estimation of the mass spectra and elution profiles of individual compounds from the superposition data is termed component resolution or deconvolution, and is especially challenging if several compounds co-elute, that is, have similar profiles in time. When the elution profiles of two or more compounds are completely co-eluting, component resolution is only possible via the simultaneous analysis of data representing different samples in which the amplitudes of the completely co-eluting compounds differ.

To date, multivariate curve resolution alternating least squares (MCR-ALS) is the only method developed for component resolution for the case of completely co-eluting compounds in mass spectrometry data. In this talk I present a newly developed alternative methodology, global analysis, which is based on a parametric model for the elution profiles. Global analysis can be shown to have better performance than MCR-ALS in terms of the estimated mass spectra for problems involving completely co-eluting components, which I will illustrate with results from a comparative study on simulated GC mass spectrometry datasets.