Benchtop (60 MHz) proton NMR spectroscopy for quantification of 16-O-methylcafestol in lipophilic extracts of ground roast coffee.

We present a method for analysing the lipophilic fraction extracted from ground coffee beans using 60 MHz proton (1H) NMR spectroscopy. In addition to the triglycerides from coffee oil, spectral features are seen from a range of secondary metabolites, such as various diterpenes. We demonstrate quantitation of a peak attributed to one such compound, 16-O-methylcafestol (16-OMC), which is of interest as a coffee species marker. It is present in low concentrations (<<50 mg/kg) in Coffea arabica L. ('Arabica') beans, but in orders of magnitude greater concentrations in other coffees, in particular the other commercially grown species C. canephora Pierre ex A. Froehner (commonly known as 'robusta'). A series of coffee extracts spiked with 16-OMC analytical standard are used to establish a calibration, and to estimate 16-OMC concentrations in a range of different coffees (Arabicas and blends with robustas). To validate the method, values obtained are compared with an analogous quantitation method that uses high field (600 MHz) NMR spectroscopy. •Quantitation of 16-O-methylcafestol in ground roast coffee extracts using benchtop (60 MHz) NMR spectroscopy•Validated by comparison with quantitative high field (600 Mz) NMR spectroscopy•Detection limit is sufficient for discovering adulteration of Arabica coffee with non-Arabica species.

Authentication of beef versus horse meat using 60 MHz 1H NMR spectroscopy.

This work reports a candidate screening protocol to distinguish beef from horse meat based upon comparison of triglyceride signatures obtained by 60 MHz (1)H NMR spectroscopy. Using a simple chloroform-based extraction, we obtained classic low-field triglyceride spectra from typically a 10 min acquisition time. Peak integration was sufficient to differentiate samples of fresh beef (76 extractions) and horse (62 extractions) using Naïve Bayes classification. Principal component analysis gave a two-dimensional "authentic" beef region (p=0.001) against which further spectra could be compared. This model was challenged using a subset of 23 freeze-thawed training samples. The outcomes indicated that storing samples by freezing does not adversely affect the analysis. Of a further collection of extractions from previously unseen samples, 90/91 beef spectra were classified as authentic, and 16/16 horse spectra as non-authentic. We conclude that 60 MHz (1)H NMR represents a feasible high-throughput approach for screening raw meat.

Electromyographic responses to prescribed mastication.

The aim was to understand between-volunteer differences in Electromyography (EMG) behaviour during chewing. EMG was used to record the electrical activity of the temporal and masseter muscles of volunteers, who carried out mastication movements by operating calibrated springs held between their incisors. The volunteers coordinated their jaw movements with the signal produced by a metronome, at four rates: 30, 60, 90 and 120 beats per minute (bpm). Raw data were analyzed to examine the distributions of the intervals between chews. For the highest prescribed chew rates, the volunteers' distributions were very similar. The distributions varied most for the 30 bpm data, suggesting that volunteers differed in their ability to carry out and maintain this prescribed chewing pattern. The data were Fourier transformed to give power spectra in the frequency domain. The low frequency (<10 Hz) region contained spectral features related to the prescribed chew rate. Principal component analysis of the power spectra revealed that readings from each volunteer clustered together, and the clusters could be largely separated. Such grouping was found irrespective of whether data from each chew rate were analyzed separately or simultaneously. This indicated that within-volunteer variance, arising from the different chew rates as well as between-session variance, is lower than between-volunteer variance; even when individuals are asked to make jaw movements in the same prescribed manner, they can nevertheless be uniquely distinguished by their muscle activity as recorded by EMG.

Detection of adulteration in cooked meat products by mid-infrared spectroscopy.

Mid-infrared spectroscopy was used to discriminate between pure beef and beef containing 20% w/w of a range of potential adulterants (heart, tripe, kidney, and liver). Spectra were acquired from raw samples and from samples cooked using two different cooking regimes. Chemometric methods (principal component analysis, partial least squares regression, and linear discriminant analysis) applied to the spectra showed that discrimination between the pure and adulterated sample types was possible, irrespective of cooking regime. The cross-validated classification success rate obtained was approximately 97%. Discrimination between all five sample types (pure beef and beef containing one of each of the four adulterants) at each level of cook was also possible, but became more difficult as the cooking level increased.

Comparison of spectroscopic techniques for the determination of Kjeldahl and ammoniacal nitrogen content of farmyard manure.

The feasibility of determining the nitrogen content of farmyard manure using infrared spectroscopy was investigated. Fifteen samples each of cattle, pig, and turkey manure were analyzed by three infrared techniques: Fourier transform mid-infrared (MIR), using attenuated total reflection (ATR); near-infrared reflectance (NIR-R); and near-infrared optothermal photoacoustic (NIR-OT). The near-infrared measurements were made at wavelengths determined respectively by four (NIR-OT) and five (NIR-R) band-pass filters. The total nitrogen (using the Kjeldahl method) and volatile (ammoniacal) nitrogen contents of all samples were measured by wet chemistry. Internally cross-validated (ICV) partial least-squares (PLS) regression was then used to obtain calibrations for the nitrogen content. The data sets obtained by each technique were treated separately. Within these sets, data from each manure type were treated both separately and combined: the best predictive ability was obtained by combining data from all three manure types. From the combined data set, the residual standard deviations and correlation coefficients for the ICV-predicted versus actual Kjeldahl nitrogen content were, respectively, 6772 mg/kg dry wt, 0.862 (MIR); 9434 mg/kg dry wt, 0.771 (NIR-OT); and 8943 mg/kg dry wt, 0.865 (NIR-R). For the ammoniacal nitrogen content, the residual standard deviations and correlation coefficients were 3869 mg/kg dry wt, 0.899 (MIR); 6079 mg/kg dry wt, 0.820 (NIR-OT); and 3498 mg/kg dry wt, 0.961 (NIR-R).

Factors affecting the release of flavor encapsulated in carbohydrate matrixes.

The effects of water content and temperature variation on the release of flavor components into the headspace over flavors, encapsulated by an extrusion process, in low water content carbohydrate matrixes is studied. The largest amounts of release occurred when the matrix was above its glass transition temperature, whether this was due to increased water content or elevated temperature. Under these conditions up to 70% of the sucrose in the matrix crystallized over a period of 10 days, as quantified using Fourier transform Raman spectroscopy. Smaller amounts of headspace release occurred when the water content of the encapsulated flavor system was decreased from 3. 5 to 3.1% w/w. Small amounts of release occurred from the "as prepared" materials, which were associated with the presence of small amounts of unencapsulated flavor oil with direct access to the headspace. It was concluded that release due to matrix permeability was relatively slow as compared with the above mechanisms.

Mid-infrared spectroscopy and chemometrics for the authentication of meat products.

Mid-infrared (MIR) spectroscopy is used to address certain issues connected with the authentication of beef and ox kidney and liver: is it possible to distinguish muscle from offal tissue; does the condition, cut of meat, or type of offal influence the distinction; can pure minced beef be distinguished from that adulterated with offal? Using partial least squares (PLS) and canonical variate analysis, predictive models are developed to identify MIR spectra of beef, kidney, and liver. Using modified SIMCA, the pure beef specimens are modeled as a single class; this model identifies spectra of unadulterated beef as such, with an acceptable error rate, while rejecting spectra of specimens containing 10-100% w/w kidney or liver. Finally, PLS regressions are performed to quantify the amount of added offal. The prediction errors obtained (+/-4.8 and +/-4.0% w/w, respectively, for the kidney and liver calibrations) are commensurate with the detection limits suggested by the SIMCA analysis.

Avoiding overfitting in the analysis of high-dimensional data with artificial neural networks (ANNs).

Complex data analysis is becoming more easily accessible to analytical chemists, including natural computation methods such as artificial neural networks (ANNs). Unfortunately, in many of these methods, inappropriate choices of model parameters can lead to overfitting. This study concerns overfitting issues in the use of ANNs to classify complex, high-dimensional data (where the number of variables far exceeds the number of specimens). We examine whether a parameter ρ, equal to the ratio of the number of observations in the training set to the number of connections in the network, can be used as an indicator to forecast overfitting. Networks possessing different ρ values were trained using as inputs either raw data or scores obtained from principal component analysis (PCA). A primary finding was that different data sets behave very differently. For data sets with either abundant or scant information related to the proposed group structure, overfitting was little influenced by ρ, whereas for intermediate cases some dependence was found, although it was not possible to specify values of ρ which prevented overfitting altogether. The use of a tuning set, to control termination of training and guard against overtraining, did not necessarily prevent overfitting from taking place. However, for data containing scant group-related information, the use of a tuning set reduced the likelihood and magnitude of overfitting, although not eliminating it entirely. For other data sets, little difference in the nature of overfitting arose from the two modes of termination. Small data sets (in terms of number of specimens) were more likely to produce overfit ANNs, as were input layers comprising large numbers of PC scores. Hence, for high-dimensional data, the use of a limited number of PC scores as inputs, a tuning set to prevent overtraining and a test set to detect and guard against overfitting are recommended.