Discrimination and classification of extra virgin olive oil using a chemometric approach based on TMS-4,4'-desmetylsterols GC(FID) fingerprints of edible vegetable oils.

A single out-line HPLC-GC (FID) analytical method is applied to acquire the chromatographic fingerprint characteristic of the TMS-4,4'-desmetylsterol derivative fraction of several marketed edible vegetable oils in order to identify and discriminate the most valuable extra-virgin olive oils from the other vegetal oils (canola, corn, grape seed, linseed, olive pomace, peanut, rapeseed, soybean, sesame, seeds (non-specified composition but usually a blend of corn and sunflower) and sunflower). The natural structure of the preprocessed data undergoes a preliminary exploration using principal component analysis and heat map-based cluster analysis. A partial least squares-discriminant model is first trained from 53 oil samples (only 3 latent variables) and externally validated from 18 test oil samples. No classification errors are found and all the test samples are correctly classified. Additional classification models are also built in order to discriminate among vegetables-oil families and excellent results have been also achieved.

Quantification of extra virgin olive oil in dressing and edible oil blends using the representative TMS-4,4'-desmethylsterols gas-chromatographic-normalized fingerprint.

This paper describes and discusses the application of trimethylsilyl (TMS)-4,4'-desmethylsterols derivatives chromatographic fingerprints (obtained from an off-line HPLC-GC-FID system) for the quantification of extra virgin olive oil in commercial vinaigrettes, dressing salad and in-house reference materials (i-HRM) using two different Partial Least Square-Regression (PLS-R) multivariate quantification methods. Different data pre-processing strategies were carried out being the whole one: (i) internal normalization; (ii) sampling based on The Nyquist Theorem; (iii) internal correlation optimized shifting, icoshift; (iv) baseline correction (v) mean centering and (vi) selecting zones. The first model corresponds to a matrix of dimensions 'n×911' variables and the second one to a matrix of dimensions 'n×431' variables. It has to be highlighted that the proposed two PLS-R models allow the quantification of extra virgin olive oil in binary blends, foodstuffs, etc., when the provided percentage is greater than 25%.

Fast-HPLC Fingerprinting to Discriminate Olive Oil from Other Edible Vegetable Oils by Multivariate Classification Methods.

A new analytical method for the differentiation of olive oil from other vegetable oils using reversed-phase LC and applying chemometric techniques was developed. A 3 cm short column was used to obtain the chromatographic fingerprint of the methyl-transesterified fraction of each vegetable oil. The chromatographic analysis took only 4 min. The multivariate classification methods used were k-nearest neighbors, partial least-squares (PLS) discriminant analysis, one-class PLS, support vector machine classification, and soft independent modeling of class analogies. The discrimination of olive oil from other vegetable edible oils was evaluated by several classification quality metrics. Several strategies for the classification of the olive oil were used: one input-class, two input-class, and pseudo two input-class.

One input-class and two input-class classifications for differentiating olive oil from other edible vegetable oils by use of the normal-phase liquid chromatography fingerprint of the methyl-transesterified fraction.

A new method for differentiation of olive oil (independently of the quality category) from other vegetable oils (canola, safflower, corn, peanut, seeds, grapeseed, palm, linseed, sesame and soybean) has been developed. The analytical procedure for chromatographic fingerprinting of the methyl-transesterified fraction of each vegetable oil, using normal-phase liquid chromatography, is described and the chemometric strategies applied and discussed. Some chemometric methods, such as k-nearest neighbours (kNN), partial least squared-discriminant analysis (PLS-DA), support vector machine classification analysis (SVM-C), and soft independent modelling of class analogies (SIMCA), were applied to build classification models. Performance of the classification was evaluated and ranked using several classification quality metrics. The discriminant analysis, based on the use of one input-class, (plus a dummy class) was applied for the first time in this study.