The corresponding roasting times and temperatures are listed in Table 1. A Shimadzu IRAffinity-1 FTIR Spectrophotometer (Shimadzu, Japan) with a DLATGS (Deuterated Triglycine Sulfate Doped with L-Alanine) detector was used in the measurements, all performed in a dry controlled atmosphere at room temperature (20 ± 0.5 °C). Diffuse reflectance (DR) measurements were performed with a Shimadzu sampling accessory (DRS8000A). Each sample was mixed with KBr and 23 mg of this mixture were placed inside

the sample port. Pure KBr was employed as reference material (background spectrum). All spectra were recorded within a range of 4000–400 cm−1 with 4 cm−1 resolution and 20 scans, and submitted to background subtraction. The spectra were also truncated to 2500 data points in the range GSK-3 beta phosphorylation mTOR inhibitor of 3200–700 cm−1, to eliminate noise readings present in the upper and lower ends of the spectra. Preliminary tests were performed in order to evaluate the effect of particle size (D < 0.15 mm; 0.15 mm < D < 0.25 mm; 0.25 mm < D < 0.35 mm) and sample/KBr mass ratio (1, 5, 10, 20 and 50 g/100 g) on the quality of the obtained spectra. The conditions that provided the best quality spectra (higher intensity and lower noise interference)

were D < 0.15 mm and 10 g/100 g sample/KBr mass ratio. In order to improve sample discrimination, the following data pretreatment techniques were evaluated: (0) no additional Oxalosuccinic acid processing (raw data), (1) mean centering, (2) normalization, (3) baseline correction employing two (3200 and 700 cm−1) or three (3200, 2000 and 700 cm−1) points, (4) first derivatives and (5) second derivatives. Mean centering was calculated by subtracting the average absorbance value of a given spectrum from each data point. Normalization was calculated by dividing the difference between the response at each data point and the minimum absorbance value by the difference between the maximum and minimum absorbance values. Such spectra pretreatments are

suggested as a means to remove redundant information and enhance sample-to-sample differences ( Wang et al., 2009). Baseline correction and derivative transformations usually compensate for baseline offset between samples and also tend to reduce instrument drift effects. Using the DR spectra (raw or normalized) and its derivatives as chemical descriptors, pattern recognition (PR) methods (PCA and LDA) were applied in order to establish whether roasted coffee husks and roasted corn could be discriminated from roasted coffee samples. For PCA analysis, data matrices were constructed so that each row corresponded to a sample and each column represented the spectra datum at a given wavenumber, after processing as previously described. LDA models were constructed based on the data that presented the best performance (group separation) in the PCA analysis.