BENEFITS OF MULTIVARIATE CURVE RESOLUTION METHODS TO ANALYZE LARGE-SCALE RAW TIME-DOMAIN NMR DATA
Abstract
Time-domain nuclear magnetic resonance (TD-NMR) provides new opportunities for large-scale and non-destructive studies of water distribution and transfers in many biological systems. TD-NMR is based on longitudinal (T1) and transverse (T2) relaxation time measurements [1]. These parameters that measure the molecular dynamics through the magnetic properties of protons (hydrogens), give access to molecular information, and to their physico-chemical properties in the investigated systems. Due to relaxation time phenomena, the measured NMR signal consists of a sum of decreasing exponentials e^((-t)⁄τ), where the distribution of τ values produces T1 or T2 spectra. Thus, the classical processing of such data consists of a Non-Negative Least Squares (NNLS) fitting procedure [2] or uses a numerical inversion of the Laplace transform [3]. Such signal processing task is known to be an ill-conditioned and ill-posed problem, resulting in a large number of solutions that small noise in the data can easily affect.
To deal with this limitation, this work aims at proposing the use of chemometrics methods as an alternative, in particular multi-curve resolution approaches [3]. When MCR is applied to TD-NMR data, the pure spectra are NMR signals that can be processed by an inverse Laplace transform without numerical instability problems. This provides reliable information on the compounds present in the system. Moreover, concentration profiles estimated by MCR allow highlighting temporal phenomena, if the system monitored is process-related, and/or increases the knowledge of spatial structures if the raw data are organised in images. The advantages of using chemometrics instead of classical processing methods will be shown through the study of changes in hydration properties of nineteen genotypes of Arabidopsis seeds during water imbibition [4] and the monitoring by Magnetic Resonance microimaging of a potato starch blend swelling process [5].