Content
Citation(s) & Footnote(s)
Book(s)No access
Paulina Wiśniewska, Ricard Boqué, Eva Borràs, Olga Busto, Waldemar Wardencki, Jacek Namieśnik, and Tomasz Dymerski. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 173 (2017): 849–853. DOI 10.1016/j.saa.2016.10.042. Online October 27, 2016; volume year 2017.
All five PDF pages read, including methods, conclusions, acknowledgments and 16 references. Figures 1–5 and Table 1 visually inspected on PDF pp. 2–5. Printed pagination equals PDF page + 848.
Design and methods
Eleven shop-purchased samples from five origin/style groups, analyzed in duplicate. Duplicates are not independent whiskies. Four analytical modes: HS-MS, FTIR transmission, FTIR ATR and UV–vis; PLS-DA classifiers. HS-MS uses the GC column for rapid volatile transfer, not resolved chromatographic identification. Headspace preparation uses 5 mL samples, 50 °C, 20-minute equilibration; authors' no-pretreatment claim should not imply instantaneous/no handling. Transmission IR uses water blank; ATR uses air background. Models use autoscaling (HS-MS), SNV plus mean centering (IR), or mean centering (UV–vis), with contiguous-block cross-validation and split 11. No independent external sample cohort is described. Exact bottle-level fold separation, number of latent variables selected, sample identities and raw spectra are not sufficiently specified here to reproduce the whole analysis.
Main finding and essential qualification
Authors argue HS-MS best separates the five classes and infer that aroma distinguishes styles. However, Table 1 (printed 851/PDF3) reports bourbon CV sensitivity 0.000 in ALL four modes, against calibration sensitivity 1.000. Bourbon CV classification-error values are 0.611, 0.639, 0.750, 0.556 respectively (HS-MS, IR-T, IR-ATR, UVV). Irish HS-MS CV sensitivity also is 0.000. Visually separated score plots are not adequate evidence of deployable classification performance. The final paragraph properly calls the study preliminary and requests more samples. Do not silently replace the table's results with the abstract's optimistic wording.
Figure interpretation
Figure 1 compares spectral fingerprints; it does not identify individual aroma-active molecules. Figure 2 HS-MS uses LV3 and LV2, not the first two axes. Figures 3–5 use different latent-variable combinations and show class proximity/convex hulls, not externally validated decision performance. Ellipses marked 95% confidence are not 95% predictive accuracy. Figure 5's low-variance third axis still participates in the visual separation; dominant variance is not synonymous with class predictive validity.
Limits and project fit
No human sensory panel, detection thresholds, preference study or consumer blend experiment. Instrumental discrimination cannot establish that human tasters perceive a corresponding difference, much less that a blend is better. Botanical origin and production method are confounded with the commercial product group; causation is not isolated. Strong bridge to ZET Authentication models are conditional on their reference populations and to existing elemental-profile studies; use as a cautionary example, not an authentication recipe.
Background statements quarantined
Introduction prints a bourbon barrel-entry condition of more than 62.5% alcohol; do not reuse as law. Generalizations about Irish triple distillation, Spanish composition, Tennessee identity and sensory equivalence are background claims and must not be promoted into authoritative profile data from this paper. No current legal advice is offered. The study's analytical contribution is distinct from these background claims.
Located evidence
