Nonetheless, not all pretreatments were satisfactory for simultaneous discrimination between roasted coffee,

roasted coffee husks and roasted corn. The spectra pretreatment steps that provided a satisfactory level of group separation when coffee and both adulterants were analyzed simultaneously were the following: (0) no additional treatment of raw data, (3) normalization with three point baseline correction and (4) first derivatives. The corresponding scatter plots obtained after Akt inhibitor PCA analysis of the data (135 samples) are displayed in Fig. 4. Roasted coffee, roasted coffee husks and roasted corn can be identified as separated groups. Roasted corn is clearly separated from the others, whereas some group overlapping is observed between coffee and coffee husks for the spectra-based plots (Fig. 4a and

b). Complete separation of the three groups was obtained after submitting the spectra to first derivatives (Fig. 4c). Evaluation of the loadings plots obtained Selleckchem 5FU after PCA analysis of raw spectra indicated that the spectral ranges that presented the highest influence on PC2 values in association with roasted corn were the following: 2250–1850 and 945–700 cm−1. In the wavenumber range 945–700 cm−1, the differences between the spectra are quite evident and they might be attributed to the presence of non-degraded starch in corn after roasting and its complete absence in roasted coffee and coffee husks (Amboni, Francisco, & Teixeira, 1999).

Differences can be also associated to the degree of saturation of the fatty acids in the triacylglycerol fraction of the coffee and corn oils (Guillén & Cabo, 1999), with the coffee oil presenting a higher degree of saturation than the corn oil (Moreau, 2002; Segall, Artz, Raslan, Jham, & Takahashi, 2005) and the correlated bands being displaced to higher wavenumbers (∼915 cm−1) than those for the corn oil. The highest influence on PC2 values in association with roasted coffee husks was observed in the range of 1600 to 1500 cm−1. In the case of normalized spectra, this website the following ranges could be associated with separation of roasted coffee: 3040–3000, 2650–2350 and 1800–1760 cm−1. Loadings obtained for first derivatives could not be associated to specific regions in the spectra. The satisfactory group separation results obtained from the principal components analysis indicate that the data should provide enough information to develop classification models for roasted coffee and each specific roasted contaminant. Thus, linear discriminant analysis (LDA) was employed in order to obtain classification models (95% confidence). LDA models were constructed employing different numbers of variables, starting with all the data points and decreasing the number of variables.