Prediction of skeletal muscle and fat mass in patients with advanced cancer using a metabolomic approach
- Cynthia Stretch
- Thomas Eastman
- Rupasri Mandal
- Roman Eisner
- David S. Wishart, Departments of Computing Science and Biology, University of Alberta
- M Mourtzakis
- C Prado
- Sambasivarao Damaraju, Cross Cancer Institute
- Ron Ball
- Russ Greiner, Dept of Computing Science; PI of AICML
- Vickie Baracos, Division of Experimental Oncology
Urine and plasma metabolites originate from endogenous metabolic pathways in different organs and exogenous sources (diet). Urine and plasma were obtained from advanced cancer patients and investigated to determine if variations in lean and fat mass, dietary intake, and energy metabolism relate to variation in metabolite profiles. Patients (n = 55) recorded their diets for 3 d and after an overnight fast they were evaluated by DXA and indirect calorimetry. Metabolites were measured by NMR and direct injection MS. Three algorithms were used [partial least squares discriminant-analysis, support vector machines (SVM), and least absolute shrinkage and selection operator] to relate patients' plasma/urine metabolic profile with their dietary/physiological assessments. Leave-one-out cross-validation and permutation testing were conducted to determine statistical validity. None of the algorithms, using 63 urine metabolites, could learn to predict variations in individual's resting energy expenditure, respiratory quotient, or their intake of total energy, fat, sugar, or carbohydrate. Urine metabolites predicted appendicular lean tissue (skeletal muscle) with excellent cross-validation accuracy (98% using SVM). Total lean tissue correlated highly with appendicular muscle (Pearson r = 0.98; P < 0.0001) and gave similar cross-validation accuracies. Fat mass was effectively predicted using the 63 urine metabolites or the 143 plasma metabolites, exclusively. In conclusion, in this population, lean and fat mass variation could be effectively predicted using urinary metabolites, suggesting a potential role for metabolomics in body composition research. Furthermore, variation in lean and fat mass potentially confounds metabolomic studies attempting to characterize diet or disease conditions. Future studies should account or correct for such variation.
Citation
C. Stretch, T. Eastman, R. Mandal, R. Eisner, D. Wishart, M. Mourtzakis, C. Prado, S. Damaraju, R. Ball, R. Greiner, V. Baracos. "Prediction of skeletal muscle and fat mass in patients with advanced cancer using a metabolomic approach". Journal of Nutrition, 142(1), pp 14-21, January 2012.| Keywords: | machine learning, bioinformatics, mass, medical informatics, metabolomics |
| Category: | In Journal |
| Web Links: | Journal |
| DOI |
BibTeX
@article{Stretch+al:JofN12,
author = {Cynthia Stretch and Thomas Eastman and Rupasri Mandal and Roman
Eisner and David S. Wishart and M Mourtzakis and C Prado and Sambasivarao
Damaraju and Ron Ball and Russ Greiner and Vickie Baracos},
title = {Prediction of skeletal muscle and fat mass in patients with advanced
cancer using a metabolomic approach},
Volume = "142",
Number = "1",
Pages = {14-21},
journal = {Journal of Nutrition},
year = 2012,
}Last Updated: February 10, 2020Submitted by Sabina P
