Computational protocol: Non-invasive diagnosis of papillary thyroid microcarcinoma: a NMR-based metabolomics approach

Similar protocols

Protocol publication

[…] For NMR spectra recorded in tissue samples, Spectra were phased, baseline corrected, and referenced to Alanine's signal at 1.47 ppm in Chenomx NMR Suite 8.1 (Chenomx Inc., Edmonton, Canada). Spectrum binning was performed on all 1H HRMAS data with a bin width of 0.04 ppm from 0.1 ppm to 10.00 ppm. In order to avoid interference caused by water suppression, the region from 4.65 ppm to 5.05 ppm was excluded. The resulting data matrix was normalized to the total area under spectrum curve. Metabolite signal assignment was performed in Chenomx NMR Suite 8.1 using a targeted profiling method with an internal database. Some ambiguous assignments were confirmed with 1H -1H COSY spectra, 1H -1H TOCSY spectra, and 1H-13C HSQC spectra. For NMR spectra recorded in plasma samples, spectra were phased, baseline corrected, and reference deconvoluted against the DSS singlet at 0 ppm using Chenomx NMR Suite 8.1. The targeted profiling method [] was used to qualify and quantify all metabolite compounds in these spectra.The normalized integral values from tissue samples and compound concentration data from plasma samples were then subjected to multivariate pattern recognition analysis using “PCA Methods” [], “PLS” [] package, and data visualization was performed using the “ggplot2” package [] in the R programing environment. Principle Component Analysis (PCA) was first used to detect grouping trends and outliers. Partial least squares discriminant analysis (PLS-DA) was then performed for class discrimination and biomarker selection. Evaluation of the PLS-DA models was performed using the goodness of-fit parameter R2 (variation in class membership explained by the model) and the predictive ability parameter Q2 (goodness of prediction, calculated by 7-fold internal cross-validation), where values of R2 and Q2 close to 1.0 represent excellent modelling. In addition, a permutation test on the response (1000 random permutations) was also computed (P<0.001 means there is no random model found with better model quality, compared with the original one). Potential biomarkers were discovered according to variable importance in the project (VIP) value and the loading plot was generated from PLS-DA analysis. A Receiver Operating Characteristic (ROC) Curve was used to build a diagnosis model by an online metabolomics analysis platform, “MetaboAnalyst 3.0” [] with the plasma metabolomics data. […]

Pipeline specifications

Software tools Chenomx NMR Suite, Ggplot2, MetaboAnalyst
Applications Miscellaneous, NMR-based metabolomics
Organisms Homo sapiens
Chemicals Alanine, Amino Acids, Cystine, Fatty Acids, Glucose, Mannose, Taurine, Tyrosine, Glutamic Acid, Pyruvic Acid, Lactic Acid, 3-Hydroxybutyric Acid