ARS | Publication request: A statistical evaluation of spectral fingerprinting methods using analysis of variance and principal component analysis

Submitted to: Food Chemistry
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 16, 2010
Publication Date: March 1, 2011
Citation: Luthria, D.L., Mukhopadhyay, S., Lin, L., Harnly, J.M. 2011. A statistical evaluation of spectral fingerprinting methods using analysis of variance and principal component analysis. Food Chemistry. 65:80-85.

Interpretive Summary: Genetics and growing conditions can lead to chemical differences in the same plant materials. A comparison of different spectral fingerprinting procedures is described that allows rapid classification of broccoli samples grown under different conditions. Spectral fingerprints were acquired for finely-powdered solid samples using Fourier transform-infrared (IR) and Fourier transform-near infrared (NIR) spectrometry and for aqueous methanol extracts of the powders (without prior separation) using molecular absorption in the ultraviolet (UV) and visible (Vis) regions and mass spectrometry with negative (MS-) and positive (MS+) ionization. The fingerprints were analyzed using nested, one-way analysis of variance (ANOVA) and principal component analysis (PCA) to statistically evaluate the quality of discrimination. All 6 methods showed statistically significant differences between the cultivars and treatments. These spectral fingerprinting tools will prove useful to analytical chemists in comparing and classifying plant materials.

Technical Abstract: Six methods were compared with respect to spectral fingerprinting of a well-characterized series of broccoli samples. Spectral fingerprints were acquired for finely-powdered solid samples using Fourier transform-infrared (IR) and Fourier transform-near infrared (NIR) spectrometry and for aqueous methanol extracts of the powders (without prior separation) using molecular absorption in the ultraviolet (UV) and visible (Vis) regions and mass spectrometry with negative (MS-) and positive (MS+) ionization. The fingerprints were analyzed using nested, one-way analysis of variance (ANOVA) and principal component analysis (PCA) to statistically evaluate the quality of discrimination. All 6 methods showed statistically significant differences between the cultivars and treatments. The significance of the statistical tests was improved by the judicious selection of spectral regions (IR and NIR), masses (MS+ and MS-), and derivatives (IR, NIR, UV, and Vis). In general, UV and VIS provided the lowest relative variance for analytical uncertainty and the largest F- and t-values for nested, one-way ANOVA and PCA, respectively.