Analysis of time course 1H NMR metabolomics data by multivariate curve resolution

• Published in: Magnetic resonance in chemistry Vol. 47; no. S1; pp. S105 - S117
• Main Authors: Karakach, Tobias K., Knight, Richard, Lenz, Eva M., Viant, Mark R., Walter, John A.
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Ltd 01-12-2009
• Subjects: Algorithms; Animals; Biological; embryogenesis; Embryonic Development; Error analysis; Fish; Fishes; Fluctuation; Least squares method; Magnetic Resonance Spectroscopy; Mathematical models; Metabolites; Metabolomics; Multivariate Analysis; multivariate curve resolution; Nephrosis - urine; nephrotoxicity; Oryzias latipes; Rats; Temporal logic; time course metabolomics; Time Factors; Urine - chemistry
• ISSN: 0749-1581; 1097-458X; 1097-458X
• DOI: 10.1002/mrc.2535

Modeling NMR‐based metabolomics data often involves linear methods such as principal component analysis (PCA) and partial least squares (PLS). These methods have the objective of describing the main variance in the data and maximum covariance between the predictor variables and some response variable respectively. If the experiment is designed to investigate temporal biological fluctuations, however, the factors obtained become difficult to interpret in a biological context. Moreover, when these methods are applied to analyze data, an implicit assumption is made that the measurement errors exhibit an iid‐normal distribution, often limiting the extent of the information recovered. A method for the linear decomposition of NMR‐based metabolomics data by multivariate curve resolution (MCR), which has been used elsewhere for time course transcriptomics applications, is introduced and implemented via a weighted alternating least squares (ALS) approach. Measurement of error information is incorporated in the modeling process, allowing the least squares projections to be performed in a maximum likelihood fashion. As a result, noise heteroscedasticity resulting from pH‐induced peak shifts can be modeled, eliminating the need for binning/bucketing. The utility of the method is demonstrated using two sets of temporal NMR metabolomics data, HgCl2‐induced nephrotoxicity in rat, and fish (Japanese medaka, Oryzias latipes) embryogenesis. Profiles extracted for the nephrotoxicity data exhibit strong correlations with metabolites consistent with temporal fluctuations in glucosuria. The concentration of metabolites such as acetate, glucose, and alanine exhibit a steady increase, which peaks at Day 3 post dose and returns to basal levels at Day 8. Other metabolites including citrate and 2‐oxoglutarate exhibit the opposite characteristics. Although the fish embryogenesis data are more complex, the profiles extracted by the algorithm display characteristics that depict temporal variation consistent with processes associated with embryogenesis. Copyright © 2009 Crown in the right of Canada. Published by John Wiley & Sons, Ltd
Multivariate curve resolution via weighted alternation squares (MCR‐wALS) has been shown to be an effective tool for modeling time course metabolomics data. Two data sets were used to demonstrate the utility of this algorithm. Concentration profiles that qualitatively describe the evolution of metabolites during the course of the experiment were obtained along with corresponding spectral profiles that can aid in identification of individual metabolites in a complex mixture of small molecules in biofluids and tissue extracts.