Prediction of Cancer-Associated Skeletal Muscle Wasting Using Targeted Profiling of Urinary Metabolites
- Roman Eisner
- Jianguo Xia
- David D. Hau
- Thomas Eastman
- Cynthia Stretch
- Sambasivarao Damaraju, Cross Cancer Institute
- Russ Greiner, Dept of Computing Science; PI of AICML
- David S. Wishart, Departments of Computing Science and Biology, University of Alberta
- Vickie Baracos, Division of Experimental Oncology
Cachexia is a complex metabolic syndrome characterized by weight loss and skeletal muscle depletion, which bodes a poor prognosis in cancer patients. While image-based methods (dual energy x-ray, computed tomography (CT)) have the specificity and precision to detect early or slowly evolving wasting, or muscle wasting without fat loss, they are costly, inconvenient and expensive. NMR-based metabolomics and statistical approaches based machine-learning potentially offer new opportunities to simplify the detection and monitoring of muscle catabolism. We obtained high-resolution 1H-NMR spectra from the urine of 96 advanced cancer patients, from which we identified and quantified 71 common urinary metabolites. We evaluated lumbar skeletal muscle area (cm2) from clinical CT images, from which we estimated the patientâs muscle change (loss, maintenance, or gain) over a period of 100 days. We applied statistical and machine-learning techniques to this labeled data to identify urinary metabolite patterns discriminating the condition of muscle loss versus muscle gain, outside a minimal margin of ï±0.75% / 100days. Several metabolites correlated with muscle loss and this may provide some biological clues to the nature of the wasting process. Â We produced a classifier based Pathway-Informed Analysis of Markov Random Fields, which robustly achieved a prediction accuracy of 79.2% for muscle loss. This method outperformed other machine learning approaches (Naive Bayes, Support Vector Machines) and Partial Least Squares Discriminant Analysis. We thus generated a simple, single-time point diagnostic urine test for cancer-associated loss of skeletal muscle. Â Our methods may be used to develop similar classifiers that can predict other metabolic conditions.
Citation
R. Eisner, J. Xia, D. Hau, T. Eastman, C. Stretch, S. Damaraju, R. Greiner, D. Wishart, V. Baracos. "Prediction of Cancer-Associated Skeletal Muscle Wasting Using Targeted Profiling of Urinary Metabolites". Metabolomics Society Meeting, August 2009.Keywords: | Machine-learning, cachexia, cancer, metabolomics, bioinformatics, NMR |
Category: | In Conference |
BibTeX
@incollection{Eisner+al:MetabolomicsSocietyMeeting09, author = {Roman Eisner and Jianguo Xia and David D. Hau and Thomas Eastman and Cynthia Stretch and Sambasivarao Damaraju and Russ Greiner and David S. Wishart and Vickie Baracos}, title = {Prediction of Cancer-Associated Skeletal Muscle Wasting Using Targeted Profiling of Urinary Metabolites}, booktitle = {Metabolomics Society Meeting}, year = 2009, }Last Updated: September 03, 2009
Submitted by Roman Eisner