Data on roasted coffee with specific defects analyzed by infrared-photoacoustic spectroscopy and chemometrics

This article contains data related to the research article entitled “Quantitative assessment of specific defects in roasted ground coffee via infrared-photoacoustic spectroscopy” (Dias et al., 2018) [1]. A method potentially able for assessing the quality of roasted ground coffees is described in the origin paper. Infrared spectroscopy and photoacoustic detection (FTIR-PAS) associated with multivariate calibration were used. The samples were obtained blending whole and healthy coffee beans (C. arabica and C. canephora) with specific blends of defects, named selections, which contain broken, sour, and black beans, skin, woods and healthy beans still not collected. In addition to a reduction in commercial value, the presence of defects compromises the sensory attributes of coffee. On the other hand, selections are commonly found in coffee crops and can be added intentionally to the product. Twenty-five selections were used to obtain a panel of 154 blends. The FTIR-PAS spectra of each sample generated the prediction model of Partial Least Squares Regression parameters, which are also presented here.

coffee crops and can be added intentionally to the product. Twenty-five selections were used to obtain a panel of 154 blends. The FTIR-PAS spectra of each sample generated the prediction model of Partial Least Squares Regression parameters, which are also presented here.
& 2018 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Subject area
Agricultural science More specific subject area Food chemistry Type of data Value of the data Data would be available as reference to agricultural and food science, notably for coffee quality researches.
Representative coffee material based on real situations of harvesting were explored. These selections composed the samples also described herein. We are not aware of the occurrence of such information in the literature.
The presented chemometric results are an interesting material for studies on multivariate statistical analysis.
Data on chemometrics applied for FTIR-PAS spectra of roasted ground coffee with several levels of quality can be useful for future studies of coffee quality monitoring. Thus, this investigation is particularly interesting to regulatory and supervisory agencies.

Data
Tables 1 and 2 detail the selections composition and the composition of the samples obtained after blending healthy coffees and the selections in different proportions. Fig. 1 presents the FTIR-PAS spectra of 154 samples featured in Table 2. Table 3 brings PLS-DA model parameters. Images of coffee defects are found in Fig. 2. The spectra data of all samples are given as supplementary material.

Samples of coffee and selections
The 25 selections differed in the proportion of specific defects and healthy coffee beans. Coffee quality specialists, bean by bean, manually picked out the whole and healthy beans, broken, sour and black beans, woods and skin of each selection ( Fig. 2 presents photographs of each group). The counting was performed based on the weight of each previously mentioned group, in percentage (Table 1). Then, the samples were built by blending a portion of each selection (20% or 40%) with each bases (100% of Arabica (i), and two mixtures of Arabica to Robusta coffees in the proportions 80:20 (ii) and 50:50 (w/w) (iii)). Robusta 100% and the bases were also evaluated ( Table 2). The final samples were roasted in a medium level of roasting (Probat Emmerich am Rhein, Germany, model PRG1Z, ERD Gas), until reaches 17% of weight loss and a luminosity, L*, between 22 and 26 (Konica Minolta portable colorimeter BC-10). After blending and roasting, the samples were ground (Ditting grinder KR805; Bachenbülach, Switzerland) on level 2.

FTIR-PAS analysis
The FTIR-PAS assessments were performed in a circular metal PAS cell of 9 mm diameter and 5 mm in depth containing the sample of RG coffee isolated from the room atmosphere with helium purging for 1 min before the analysis, reducing the water vapor and carbon dioxide in the sample chamber. After infrared light had been focused on sample, an ultrasensitive microphone detected the PAS signal. The resulted PAS spectrum was an average of 16 scans, with 4 cm À 1 resolution in a wavenumber region of 600-4000 cm À 1 . Before analysis, the PAS signal was calibrated with a polyethylene standard sample. The PAS signal normalization process to a black body coated reference sample, commonly named as carbonblack, was performed for eliminating the influence of the non-uniform intensity of the light source spectrum. Thus, the FTIR-PAS normalized signal was the ratio between the sample PAS signal amplitude numeral are the identification of the sample basis (Arabica coffee, A; Arabica/Robusta 80:20 w/w blend, AR20, or 50:50, AR50), and the last numeral is the selection identification (#1 -25, from Table 1). For example, 20AR50_1 is the sample containing 20% of the selection #1 in the Arabica/Robusta 50:50 basis. b The sum of whole and healthy beans from basis and of whole and healthy already included in selection (%).  Table 2, and data spectra is given in Supplementary material). and the carbon-black PAS signal amplitude. The resulting signal, which generates the PAS spectrum, is dependent on the composition of the sample because it is directly proportional to the amount of light energy absorbed by the sample at each wavenumber.

Multivariate data analysis
PLS-DA (Partial Least Squares Regression -Discriminant Analysis) was applied for the set of PAS spectra of coffee samples. The optimum PLS-DA model dimensions were determined by the minimum RMSECV value for the calibration samples, obtained by the leave-one-out procedure with 108 samples. This procedure resulted in the choice of six latent variables for mean-centered model development. Table 3 presents specific parameters of this data analysis.