FTIR and Chemometrics as Effective Tools in Predicting the Quality of Specialty Coffees

Abstract

Cup tasting is the most important tool to access the quality of coffee beans. However, the use of sensory evaluation alone can present some problems, since bias from the previous knowledge of a particular sample and health conditions of the taster can influence the results. Given the well-established potential of spectroscopic methods in coffee quality evaluation, in the present study, we sought to evaluate the potential of FTIR spectroscopy for quantitative evaluation of specialty coffee quality. Samples of specialty coffee were provided by the Federação dos Cafeicultores do Cerrado Mineiro and Fazenda Barinas. They were roasted in IKAWA coffee roaster, analyzed by a group of Q-graders, and submitted to FTIR analysis. Physicochemical analyses (pH, titratable acidity, brix, total solids, and browning compounds) were also employed to show potential differences. Only pH showed significant difference between the beverages. PLS results showed consistent models for predicting the quality previously given by the cuppers, with low values of RMSEC and RMSEP (0.23 both). Also, the models showed high values of Rc (0.99) and Rv (0.97). The whole spectra were considered as important to classify the coffees by their quality, showing the complexity of the beverage.

Full text

FTIR and Chemometrics as Effective Tools in Predicting the Quality
of Specialty Coffees
Verônica Belchior
1
&Bruno Gonçalves Botelho
2
&Susana Casal
3
&Leandro S. Oliveira
1,4
&Adriana S. Franca
1,4
Received: 4 April 2019 /Accepted: 26 July 2019
#Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract
Cup tasting is the most important tool to access the quality of coffee beans. However, the use of sensory evaluation alone can
present some problems, since bias from the previous knowledge of a particular sample and health conditions of the taster can
influence the results. Given the well-established potential of spectroscopic methods in coffee quality evaluation, in the present
study, we sought to evaluate the potential of FTIR spectroscopy for quantitative evaluation of specialty coffee quality. Samples of
specialty coffee were provided by the Federação dos Cafeicultores do Cerrado Mineiro and Fazenda Barinas. They were roasted
in IKAWA coffee roaster, analyzed by a group of Q-graders, and submitted to FTIR analysis. Physicochemical analyses (pH,
titratable acidity, brix, total solids, and browning compounds) were also employed to show potential differences. Only pH
showed significant difference between the beverages. PLS results showed consistent models for predicting the quality previously
given by the cuppers, with low values of RMSEC and RMSEP (0.23 both). Also, the models showed high values of Rc (0.99) and
Rv (0.97). The whole spectra were considered as important to classify the coffees by their quality, showing the complexity of the
beverage.
Keywords Cup quality .Chemometrics .Partial least square regression (PLS) .Attenuated total reflectance (ATR) .Fourier
transform infrared spectroscopy (FTIR) .Specialty Coffee Association (SCA)
Introduction
The quality of a cup of coffee begins to be defined when the
plant starts to develop. The selected varieties, harvesting and
post-harvesting methods applied, and later the roasting pro-
cess followed by the elaboration of blends are factors that will
greatly influence the final quality of the beverage. The delicate
flavor of a cup of coffee is the final expression of a great chain
of chemical and physical transformations that link the seed to
the cup (Di Donfrancesco etal. 2014; Sunaharum et al., 2014).
Cup tasting is the most important and common tool to
access the quality of green coffee (Craig et al. 2018; Tolessa
et al. 2016). Not only coffee but also tea, perfume, and tobacco
industries often use “experts”for this evaluation. The coffee
experts, also called “cuppers,”are professionals that accumu-
late years of wide knowledge about this product (Di
Donfrancesco et al. 2014). Usually, different producing coun-
tries have their methods for evaluating coffee (Craig et al.
2018; Santos et al. 2012). But among the available methods
for sensory analysis of coffee, those of the “Specialty Coffee
Association”(SCA) classification are considered the most
suitable for high-quality coffees due to their recommended
use of a specific protocol to carry out sensory analysis.
These protocols are based on objective assessment methods,
such as the presence or absence of sweetness and defects, thus
minimizing subjectivity in comparison with other methodolo-
gies (Leloup et al. 2004).
However, the use of the experts can present some prob-
lems. Bias from the preference and knowledge of a particular
sample, the influence of external factors, the specific health
conditionsof the taster, and changes in the personal abilities of
evaluation can affect the results. These issues encourage the
use of alternative evaluation tools (Di Donfrancesco et al.
*Verônica Belchior
belchior.veronica@gmail.com
1
PPGCA, Universidade Federal de Minas Gerais, Av. Antônio Carlos,
6627, Belo Horizonte, MG 31270-901, Brazil
2
DQ, Universidade Federal de Minas Gerais, Av. Antônio Carlos,
6627, Belo Horizonte, MG 31270-901, Brazil
3
LAQV/REQUIMTE, Faculdade de Farmácia, Universidade do
Porto, R. Jorge de Viterbo Ferreira 228, 4050-313 Porto, Portugal
4
DEMEC, Universidade Federal de Minas Gerais, Av. Antônio
Carlos, 6627, Belo Horizonte, MG 31270-901, Brazil
Food Analytical Methods
https://doi.org/10.1007/s12161-019-01619-z

2014; Lindinger et al. 2008). Sensory-based evaluation can
also be considered a time-consuming and sensitive assessment
regarding the presence of a well-trained professional. Recent
studies have demonstrated the potential of spectroscopy-based
methods for establishing rapid parameters of quality in the
analysis of food, with mid (FTIR) and near (NIR) infrared
being the most commonly employed techniques (Franca and
Nollet 2017).
Recent studies have demonstrated the potential of FTIR in
coffee analysis (Barbin et al. 2014). Applications include dis-
crimination between coffee species and varieties (Wang et al.
2011), determining adulterations in roasted and ground cof-
fees (Reis et al. 2013a,b;Reisetal.2016), and discrimination
of defective coffee beans (Craig et al. 2012; Craig et al. 2014,
2015). The promising results obtained with discrimination and
quantification of low-quality (defective) coffees, which affect
significantly the sensory quality of the beverage, indicated that
this technique could be associated to sensory evaluation. In
previous studies, we developed classification models that
were able to provide qualitative discrimination between
espresso, based on generic descriptions (intensity and a few
sensory parameters) provided by the manufacturers and also
on sensory characteristics established by a sensory panel
(Belchior et al. 2019; Belchior et al. 2016). Models were also
developed for discrimination of coffees classified by cup qual-
ity according to the Brazilian legislation (soft, hard or hardish,
rioysh, rio, and rio zona) (Craig et al. 2018).
It is clear from these studies that FTIR is a promising tech-
nique in coffee quality evaluation. However, in the aforemen-
tioned studies, only qualitative discrimination was attempted.
Therefore, in the present study, we sought to confirm the
potential of FTIR also for quantitative evaluation of sensory
characteristics of specialty coffees. Physicochemical analyses
were also employed to show whether the beverages would be
different by their characteristics. Partial least squares regres-
sion (PLS) was employed to construct models able to predict
and establish a sensory profile based on the score of quality
given for specialty coffee according to SCA.
Materials and Methods
Materials
A total of 28 green coffee bean samples were provided by
Federação dos Cafeicultores do Cerrado Mineiro (Patrocínio,
Minas Gerais, Brazil) and Fazenda Barinas (Araxá, Minas
Gerais, Brazil). They represented arabica coffee from the be-
ginning of the 2016 crop, submitted to two common post-
harvesting processing methods in Brazil: dry (natural coffee)
and wet (pulped natural coffee), which were known to be
specialty coffee based on previous evaluations.
Roasting and Sensory Analysis
The samples were roasted in an IKAWA®Sample
Roaster Pro (London, UK) supplied by Macchine Per
Caffè Ltda (São Paulo, São Paulo, Brazil). All samples
were submitted to the same roasting profile developed
in accordance with the SCA (Specialty Coffee
Association) protocol for sensorial analysis. The roasts
weretakenin4′34″minutes for each 50 g batches of
green coffee. The temperature of the roaster ranged
from 170 to 227 °C. All the samples were roasted in
duplicate (totaling 56 roasts) in order to assure differ-
ences between the roasts from the equipment itself and
inherent to the chemical composition of the beans.
Roasted samples were ground in the Porlex Mini®
(Porlex Grinders, Osaka, Japan) using the finer regulation
to obtain more homogeneous samples (D< 0.150 mm)
andthenanalyzedbysixprofessionalQ-graders accord-
ing to the SCA protocol for sensory analysis of coffee.
The coffees were classified according to the quality of the
beverage given by the means of global scores and aromat-
ic descriptors. The samples assigned to this work are con-
sidered specialty coffees with scores ranging from 81 to
91 points (Table 1).
Physicochemical Analyses
The physicochemical analyses employed included pH, titrat-
able acidity, brownish compounds, Brix degree, and total
solids. For those analyses, two aliquots of beverage were tak-
en and analyzed in duplicate, totaling 224 results (56 bever-
ages × 2 aliquots × 2 analyses).
pH and Titratable Acidity
The pH analysis was conducted in the pHmeter Crison
Basic 20+ (Barcelona, Spain) at room temperature. For
this purpose, 40 mL of beverage was used. The titratable
acidity measurement was performed according to the
methodology adopted by Gloess et al. (2014)withbever-
agetitratedwith0.1mol/LNaOHsolutionuptopH8.0.
Browning Compounds
The analysis was conducted according to the methodol-
ogy proposed by Lopez-Galilea et al. (2007). A 1:40
dilution with deionized water was prepared and the
brownish compounds were evaluated at 420 nm in
UV-visible spectrophotometer UV300 Unicam
(England, UK).
Food Anal. Methods

Table 1 Characteristics of the samples provided by the Federation of Coffee Growers of Cerrado Mineiro and Fazenda Barinas
Sample Sensorial Score (mean ± SD)* City Process Altitude (m) Variety Sensory description
A1 91.2 ± 1.0 Serra do Salitre Natural 1010 Red Catuai Floral, sweet notes, caramel, cocoa, and red fruits
A2 91.0 ± 1.0 Araxá Natural 960 Topázio MG1190 Jabuticaba liqueur, cognac, malty, dark chocolate, blackberry, grape, and blackcurrant
A3 90.4 ± 1.5 Coromandel Natural 1080 Yellow Bourbon Sweet notes of caramel, tropical fruits, and spicy
A4 88.7 ± 1.0 Patos de Minas Pulped natural 1250 IAC 125 RN Sweet and caramel, chocolate notes
A5 88.4 ± 1.0 Patrocínio Pulped natural 1000 Acaiá do Cerrado Milk chocolate, caramel, buttery
A6 88.2 ± 1.0 Presidente Olegário Pulped natural 1080 Yellow Bourbon Cocoa notes, red fruits, almonds, caramel, and dulce de leche
A7 87.7 ± 2.0 Patos de Minas Natural 1000 Red Catuai Milk chocolate, red fruits
A8 87.6 ± 1.6 Patrocínio Natural 1020 Red Bourbon Fruity, high acidity, caramel
A9 87.6 ± 2.5 Campos Altos Natural 1190 Topázio MG1190 Red fruits, milk chocolate
A10 87.2 ± 1.5 Patrocíio Natural 1020 Yellow Bourbon Sweet aroma, milk chocolate
A11 87.0 ± 1.5 Araxá Natural 960 Red Catuai Caramel notes, dulce de leche, almonds
A12 87.0 ± 1.2 Araxá Natural 960 Topázio Fruity, passionfruit, mango, honey
A13 86.0 ± 1.0 Campos Altos Pulped natural 1190 Yellow Bourbon Chocolate ao leite, castanhas, caramelo
A14 86.0 ± 1.5 Araxá Natural 960 Yellow Bourbon Milk chocolate, caramel, creamy, clean, sweet, and slightly notes of mango
A15 86.0 ± 1.5 Araxá Natural 960 Red Catuai Chocolate, caramel, blackberry, jabuticaba liqueur
A16 86.0 ± 1.2 Araxá Natural 960 Red Catuai Vanilla, chocolate, slightly fruity, notes of passion fruit
A17 83.0 ± 0.8 Araxá Natural 960 Topázio Chocolate, hazelnuts
A18 82.0 ± 0.1 Campos Altos Natural 980 Yellow Catuai Nuts, caramel
A19 82.0 ± 0.3 Carmo do Paranaíba Natural 1170 Red Catuai Nuts, caramel
A20 82.0 ± 0.2 Patrocínio Natural 1116 Mundo Novo Nuts, caramel
A21 82.0 ± 0.3 Patos de Minas Natural 1003 Red Catuai Nuts, caramel
A22 82.0 ± 0.3 Rio Paranaíba Natural 1200 Yellow Bourbon Nuts, caramel
A23 82.0 ± 0.4 Coromandel Natural 920 Red Catuai Nuts, caramel
A24 82.0 ± 0.5 Patrocínio Natural 1128 Red Catuai Nuts, caramel
A25 81.0 ± 0.2 Patrocíio Natural 920 Yellow Catuai Nuts, caramel, short aftertaste
A26 81.0 ± 0.2 Presidente Olegário Natural 1080 Red Catuai Nuts, caramel
A27 81.0 ± 0.3 Rio Paranaíba Natural 1080 Red Catuai Nuts, caramel
A28 81.0 ± 0.4 Coromandel Natural 920 Red Catuai Nuts, caramel
Food Anal. Methods

Brix and Total Solids
Beverages were directly analyzed for the Brix values on a
VWR Refractometer Reader (0–54 Bx/1.33–1.42 RI). The
total solid analysis followed Gloess et al. (2014), with a 10-
g portion of coffee beverage being dried at 105 °C until
constant weight.
FTIR Analysis
The 56 samples of roasted coffee were submitted to FTIR
analysis employing a Shimadzu IRAffinity-1 FTIR
Spectrophotometer (Shimadzu, Japan) with a DLATGS
(deuterated triglycine sulfate doped with L-alanine) detec-
tor. Two fractions were withdrawn from each sample and
analyzed in duplicate, totaling 224 spectra (56 beverages ×
2 aliquots × 2 analyses).
Data Processing and Statistical Analysis
One-way ANOVA at a confidence level of α=0.05was
applied to the data from the physicochemical analysis, by
MINITAB® software version 17.1.0, 2013.
The comparison of the coffee quality classification by
the SCA score and the other analyses, physicochemical
and spectral data, was performed with MATLAB® soft-
ware version 7.9, 2009 (The MathWorks, Natick, MA,
USA) and PLS Toolbox® 6.7.1, 2012 (Eigenvector
Technologies, Manson, WA). PLS models were built to
predict the sensory analysis results and the variables re-
lated to them using the ATR spectra as chemical descrip-
tors. For this purpose, all the 224 spectra were divided
into calibration (157 samples) and validation (67 samples)
sets based upon the Kennard-Stone algorithm. To reduce
the effect of noise, remove redundant information, and
enhance sample-to-sample differences, the following data
pre-processing techniques were applied to the obtained:
orthogonal signal correction (OSC), standard normal var-
iation (SNV), and mean center (MC).
The number of latent variables in the model was
chosen based on the lowest value of RMSECV (root
mean square error of cross-validation) obtained by
“Random Subsets”cross-validation. Model performance
was measured by evaluating the root mean square errors
for both calibration (RMSEC) and validation (RMSEP)
sets, calculated as follows:
RMSEC ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∑Ic
i¼1yi−^
yi

2=Ic
rð1Þ
RMSEP ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∑IP
i¼1yi−^
yi

2=Ip
rð2Þ
where y
i
and ŷ
i
correspond to the real and predicted
adulteration levels of sample i,andI
C
and I
P
are the total
number of samples in the calibration and prediction
(validation) sets, respectively. The models with better pre-
diction ability present lower values of RMSEC and
RMSEP (Reis et al. 2016). The calibration and validation
correlation coefficients between the reference values and
the values predicted by the model, Rc and Rv, respective-
ly, were also used to assess the ability of the models.
Results and Discussion
FTIR Analysis
The spectra refer to the average of the samples grouped into
four classes according to their score of sensorial quality: 81–
83, 84–86, 87–89, 90+, as presented in Fig. 1a. The full spec-
tra of roasted coffee submitted to the OSC, SNV, and mean
center are also presented in Fig. 1b. The two bands at the
2900–2850 cm
−1
inferred the influences of the vibrations of
the C–H bonds of the caffeine and lipid molecules (Barbosa,
2007; Craig et al. 2012). The marked 1750 cm
−1
is attributed
to carbonyl (C = O) vibration (Barbosa, 2007; Pavia et al.
2010). According to Pavia et al. (2010), this region is allocated
to absorbing molecules of anhydride and esters and is very
close to the region 1725–1700 cm
−1
that is assigned to the
vibration of the carbonyl group (C = O) of carboxylic acids
and ketones. Carboxylic acids are known to contribute to the
acidity of the coffee (Estaban-Díez et al., 2004; Ribeiro et al.
2011), while ketones contribute to aromas with characteristics
like woody, cucumber, cooked fruit, and almonds (Wang and
Linn, 2012). Wang et al. (2009) identified the region as vibra-
tion of aliphatic acid molecules relating to the aromatic coffee
profile. The presence of other carbonyl compounds in the
range of 1800–1680 cm
−1
as aldehydes is related to specific
organoleptic properties of coffee (Wang et al. 2009).
Figure 1also shows bands in the region 1650–1600 cm
−1
,
which have been associated to caffeine absorption (Lyman
et al. 2003; Paradkar and Irudayaraj, 2002;Wangetal.
2009) and employed in predictive models for quantitative
analysis of caffeine. Pavia et al. (2010) reported the region
between 1680 and 1630 cm
−1
as the carbonyl amide group
vibrations. Trigonelline is another substance also found in
coffee that generates absorbance peaks in that region.
During the roasting process, trigonelline is decomposed in-
to pyridines and pyrroles. According to Illy and Viani
(1998), pyridines are responsible for the characteristic aro-
ma of roasted coffee.
Several bands can be observed at the 1360 cm
−1
and 1230–
1000 cm
−1
regions. Carbohydrates generally have absorption
bands in the region between 1400 and 900 cm
−1
, the so called
Food Anal. Methods

“fingerprint region,”which is responsible for most of the spec-
tral bands (Kemsley et al., 1995a,b). Pavia et al. (2010)de-
scribe the same region as responsible for vibrations of the C–
O group of alcohols, ethers, esters, carboxylic acids, and an-
hydrides. According to Silverstein and Webster (1998), this
region is a characteristic of the vibration of bonds C–H, C–O,
and C–N. However, accurate chemical assignments in this
region of the spectra are considered a challenge due to highly
coupled vibration modes of polysaccharide backbones (Craig
et al. 2018). Chlorogenic acids are a family of esters formed
by quinic acid, and of one to four residues of caffeic acid, p-
coumaric acid, and ferulic acid. The spectral region between
1450 and 1150 cm
−1
is characteristic of chlorogenic acids
(Craig et al. 2012,2015; Lyman et al. 2003;Paviaetal.
2010;J.Wangetal.2009). The band 930 cm
−1
is attributed
to the presence of residues of 3,6-anhydro-galactopyranose
(Gomez-Ordoñez and Rupérez, 2011) that may result from
the thermal degradation of polysaccharides such as
arabinogalactans or galactomannan. Franca et al. (2005)indi-
cate that carbohydrates, trigonelline, and chlorogenic acid
a
b
Fig. 1 aFull spectra (3000–900
cm
−1
) of roasted coffee in original
scale. bFull spectra of roasted
coffee submitted to OSC, SNV,
and mean center
Table 2 Values of Fand pof the physicochemical parameters analyzed
for the coffee beverage
Parameters Fp
Brix (%) 1.89 0.12
Total solids (g/100 mL) 0.25 0.96
pH 3.97 0.02
Titratable acidity (mmol/100 mL) 0.78 0.60
Browning compounds (*abs 420 nm) 0.90 0.50
The data italicized is the significant one, according to the pvalue
α=0.05
*abs.: spectrophotometer absorbance UV-vis
Food Anal. Methods

levels can decrease after roasting, so variations that are expect-
ed in the chemical composition can affect the spectrum in the
regions between 1700 and 600 cm
−1
(Craig et al. 2012).
Physicochemical Analysis
The one-way ANOVA test results for the physicochemical
analyses did not show a significant difference between the
parameters, except for the pH results (Table 2). The pH values
were in accordance with the literature (Gloess et al. 2013;
Lopez-Galilea et al., 2007). The results in Fig. 2indicate that
pH values tend to decrease as the beverage score increases.
According to SCA (2015), the perceived acidity in coffee is
considered a positive quality attribute of the beverage.
However, pH alone is not sufficient to explain the quality of
the perceived acidity in coffees. Folmer (2017)andLopez-
Galilea et al. (2007) discuss the relevance of both parameters,
titratable acidity and pH, for the perception and quality of the
acidity of specialty coffees, suggesting the results may reflect
a superior quality of the coffees with higher score. No differ-
ences related to the titratable acidity may reflect the concen-
tration of other acids that are not related to the quality but that
may influence the perceived acidity, like chlorogenic acids
(Sunaharum et al., 2014).
Partial Least Squares Regression
Figure 3shows the experimental versus estimated values ob-
tained for the model built with the spectra and quality score
given by the Q-graders. The model was optimized by the
outlier removal (Table 3) and the chosen model was built with
2 latent variables, which explained 81.2% of the accumulated
variance of the spectrum data and 99.71% of the score data.
The RMSEP and RMSEC values were 0.23% and 0.23%,
respectively, and the coefficients of the calibration and valida-
tion correlation between both the spectra and score data were
Fig. 2 pH average of the
beverages classified by scores
Fig. 3 Experimental versus
predicted values by the PLS
models of the data submitted to
the OSC, SNV, and mean center
Food Anal. Methods

0.99 and 0.97, respectively. The values for external validation
were also included in the models, showing that many of these
are in the same range of the estimated values for the samples.
As it can be noticed in the figure, the models provided good
correlation between the experimental and predicted data.
Thus, it is possible to say the coffees with different scores of
qualities are separated by this characteristic.
Figure 4shows the VIP scores of the model and the marked
peaks are the ones that were the most important for the result.
It is possible to observe that the whole spectra were important
for the determination of the coffee classification correlated to
the SCA scores. It is possible to observe that the entire spec-
trum was important for the determination of the coffee classi-
fication correlated to the SCA scores. This is related to the fact
that the several substances that affect the coffee sensory pro-
file absorb throughout the whole spectrum. Furthermore, sen-
sory parameters are affected by a combination of various
chemical compounds at the same time, so it is not possible
to single out a specific absorbance region. In our previous
study on sensory profiles of espresso coffees (Belchior et al.
2019), we also observed high VIP scores all over the spectra.
Since the whole spectrum was relevant for the classification
of the samples in this study, it is important to notice the influ-
ence of all of the chemical compounds created by the roasting
in the beverage. Coffee is a very complex matrix and suscepti-
ble to the variables of cultivation, harvesting, post-harvesting,
storage, roasting, grinding, and extraction. Many aspects are
related to the sensorial variations perceived by the tasters
(Belchior et al. 2019). Thus, the aromatic profile, flavor, after-
taste, acidity, body, and balance are considered the attributes
that mostly influence the perception of the beverage (Belchior
et al. 2019). In addition, the samples need to be roasted 24 h
prior to the sensory analysis. Although this procedure provides
a certain consistency predicted by the protocols of the SCA, the
roasting profile can vary because of the environmental changes,
both internally and externally to the roaster (Folmer 2017;Wei
and Tanokura 2015). Also, the difference inherent to the beans
needs to be considered, like density and moisture.
Folmer (2017) says the whole time of roasting can be
divided into three phases: drying, Maillard reaction, and
development. The final time of roasting determines the
sensory profile of coffee in terms of composition of
organic acids, chlorogenic acid derivatives, sugar
caramelization, volatile composition (whether positive
or negative), lipid migration, and melanoidin production,
composition related to the main attributes evaluated in
the sensory analysis (Folmer 2017; Sunaharum et al.,
2014;Bhumiratanaetal.2011; Buffo and Cardelli-
Freire 2004).
Table 3 Optimization of
the PLS model through
the outliers detection and
removal (final model in
bold)
Model 1° 2°
Calibration set 150 149
Validation set 74 67
Latent variables 2 2
RMSEC 0.24 0.23
RMSEP 0.99 0.23
Rc 0.99 0.99
Rv 0.94 0.97
VL, latent variables; RMSEC, root mean
square error of calibration; RMSEP, root
mean square error of validation; Rc,cali-
bration correlation; Rv, validation
correlation
Fig. 4 VIP scores of the PLS models of the data submitted to the OSC, SNV, and centralization on average
Food Anal. Methods

In this case, proposals for validation of the sensory analysis
performed by Q-graders and roasting profile are relevant. The
results are promising for the classification of specialty coffees
and show the relevance of FTIR as a fast and efficient alter-
native for the proposed objective.
Conclusion
ATR-FTIR was shown to be a reliable tool for predicting
sensory attributes of roasted coffee samples. The results of
the physical-chemical analyses presented statistical differ-
ences only for pH values. However, the results were promis-
ing from the standpoint of chemometrics, with robust models
with high correlation coefficients of calibration and validation,
especially for the spectral data of FTIR. These results were
ideal for predicting the classification of specialty coffees ac-
cording to their score given by the cuppers, even with a greater
correlation. The analysis of the whole spectra rather than the
physicochemical characteristics of the coffee may be more
efficient and interesting from an industrial and scientific point
of view, given the complexity of the coffee itself.
Conflict of Interest Statement Verônica Belchior declares that she has
no conflict of interest. Bruno Gonçalves Botelho declares that he has no
conflict of interest. Susana Casal declares that she has no conflict of
interest. Leandro S. Oliveira declares that he has no conflict of interest.
Adriana S. Franca declares that she has no conflict of interest.
Informed Consent Informed consent was obtained from all individual
participants included in the study.
Funding Information This work was financially supported by the
Brazilian Government Agency CNPq and CAPES. Part of this work
was developed at the University of Porto, supported by the European
Union (FEDER funds POCI/01/0145/FEDER/007265) and National
Funds (FCT/MEC) under the Partnership Agreement PT2020 UID/
QUI/50006/2013.
Compliance with Ethical Standards
All procedures performed in studies involving human participants were
submitted and approved by the Ethics Committee on Research with
Human Subjects (CAAE, UNIBH, Belo Horizonte, Minas Gerais,
Brazil, 56961316.0.0000.5093).
References
Barbin DF, Felicio ALSM, Sun DW, Nixdorf SL, Hirooka EY (2014)
Application of infrared spectral techniques on quality and composi-
tional attributes of coffee: an overview. Food Res Int 61:23–32.
https://doi.org/10.1016/j.foodres.2014.01.005
Barbosa LCA (2007) Espectrocopia no Infravermelho na caracterização
de compostos orgânicos. UFV, Viçosa
Belchior V, Botelho BG, Oliveira LS, Franca AS (2019) Attenuated total
reflectance Fourier transform spectroscopy (ATR-FTIR) and
chemometrics for discrimination of espresso coffees with different
sensory characteristics. Food Chem 273:178–185. https://doi.org/
10.1016/j.foodchem.2017.12.026
Belchior V, Franca AS, Oliveira LS (2016) Potential of diffuse reflectance
infrared Fourier transform spectroscopy and chemometrics for cof-
fee quality evaluation. Int J Food Eng 2:1):1–1):8. https://doi.org/10.
18178/ijfe.2.1.1-8
Bhumiratana E, Adhikari K, Chambers E IV (2011) Evolution of sensory
aroma attributes from coffee beans to brewed coffee. J Food Sci
Technol 44(10):2185–2192. https://doi.org/10.1016/j.lwt.2011.07.
001
Buffo RA, Cardelli-Freire C (2004) Coffee flavour: an overview. Flavour
and Fragrance Journal 19(2):99–104. https://doi.org/10.1002/ffj.
1325
Craig AP, Franca AS, Oliveira LS (2012) Discrimination between defec-
tive and non-defective roasted coffees by diffuse reflectance infrared
Fourier transform spectroscopy. Lwt-Food Science and Technology
47(2):505–511. https://doi.org/10.1016/j.lwt.2012.02.016
Craig AP, Franca AS, Oliveira LS, Irudayaraj J, Ileleji K (2014)
Application of elastic net and infrared spectroscopy in the discrim-
ination between defective and non-defective roasted coffees. Talanta
128:393–400. https://doi.org/10.1016/j.talanta.2014.05.001
Craig AP, Franca AS, Oliveria LS, Irudayaraj J, Ileleji K (2015) Fourier
transform infrared spectroscopy and near infrared spectroscopy for
the quantification of defects in roasted coffees. Talanta 134:379–386
Craig AP, Botelho BG, Oliveira LS, Franca AS (2018) Mid infrared
spectroscopy and chemometrics as tools for the classification of
roasted coffees by cup quality. Food Chem 245:1052–1061.
https://doi.org/10.1016/j.foodchem.2017.11.066
di Donfrancesco B, Gutierrez Guzman N, Chambres E IV (2014)
Comparison of results from cupping and descriptive sensory analy-
sis of Colombian brewed coffee. J Sens Stud 29(4):301–311. https://
doi.org/10.1111/joss.12104
Estaban-Díez I, Gonzalez-Saiz JM, Pizarro C (2004) Prediction of senso-
ry properties of espresso from roasted coffee samples by near infra-
red spectroscopy. Anal Chim Acta 525(2):171–182. https://doi.org/
10.1016/j.aca.2004.08.057
Folmer B (2017) The craft and science of coffee. Academic Press
Elsevier, London
Franca AS, Mendonça JCF, Oliveira SD (2005) Composition of green
and roasted coffees of different cup qualities. LWT-Food Science
and Technology 38:709–715. https://doi.org/10.1016/j.lwt.2004.
08.014
Franca AS, Nollet LML (2017) Spectroscopic methods in food analysis.
CRC Press, Boca Raton
Gloess AN, Schönbächler B, Klopprogge B, D’Ambrosio L, Chatelain K,
Bongartz A, Strittmatter A, Rast M, Yeretzian C (2013) Comparison
of nine common coffee extraction methods: Instrumental and senso-
ry analysis. Eur Food Res Technol 236(4):607–627. https://doi.org/
10.1007/s00217-013-1917-x
Gloess AN, Vietri A, Wieland F, Smrke S, Scönbächler B, López JAS,
Petrozzi S, Bongers S, Koziorowski B, Yeretzian C (2014) Evidence
of different flavour formation dynamics by roasting coffee from
different origins: on-line analysis with PTR-ToF-MS. Int J Mass
Spectrom 365–366:324–337. https://doi.org/10.1016/j.ijms.2014.
02.010
Gomez-Ordoñez E, Rupérez P (2011) FTIR-ATR spectroscopy as a tool
for polysaccharide identification in edible brown and red seaweeds.
Food Hydrocoll 25(6):1514–1520. https://doi.org/10.1016/j.
foodhyd.2011.02.009
Illy A, Viani R (1998) Espresso coffee: the chemistry of quality (ed)
Academic Press Inc, San Diego, pp253. In: Siegler BC (ed) Food
Engineering. Nova Science Publishers, New York, pp 1–47
Kemsley EK, Ruault S, Wilson RH (1995a) Discrimination between
Coffea arabica and Coffea canephora variant Robusta beans using
infrared spectroscopy. Food Chem 54:321–326
Food Anal. Methods

Kemsley EK, Ruault S, Wilson RH (1995b) Discrimination between
Coffea arabica and Coffea canephora variant Robusta beans using
infrared spectroscopy. Food Chem 54:321–326
Leloup V, Gancel C, Liardon R, Rytz A, Pithon A (2004) Impact of wet
and dry process in green coffee composition and sensory character-
istics. In: International Conference in Coffee Science. ASIC,
Bangalore
Lindinger C, Labbe D, Pollien P, Rytz A, Juillerat MA, Yeretzian C,
Blank I (2008) When machine tastes coffee: instrumental approach
to predict the sensory profile of espresso coffee. Anal Chem 80(5):
1574–1581. https://doi.org/10.1021/ac702196z
Lopez-Galilea I, De Peña MP, Cid C (2007) Correlation of selected con-
stituents with the total antioxidant capacity of coffee beverages:
Influence of the brewing procedure. J Agric Food Chem 55(15):
6110–6117. https://doi.org/10.1021/jf0209793
Lyman DJ, Benck R, Dell S, Merle S, Murray-Wigelath J (2003) FTIR-
ATR analysis of brewed coffee: effect of roasting conditions. J Agric
Food Chem 51(11):3268–3272. https://doi.org/10.1021/jf0209793
Paradkar MM, Irudayaraj J (2002) Rapid determination of caffeine con-
tent in soft drinks using FTIR-ATR spectroscopy. J Food Chem 78:
261–266. https://doi.org/10.1016/S0308-8146(02)00116-4
Pavia DL, Lampan GM, Kriz GS, Vyvyan JR (2010) Introdução à
espectroscopia. Cengage Learning, Washington
Reis N, Franca AS, Oliveira LS (2013a) Performance of diffuse reflec-
tance infrared Fourier transform spectroscopy and chemometrics for
detection of multiple adulterants in roasted and ground coffee. LWT-
Food Science and Technology 53(2):395–401. https://doi.org/10.
1016/j.lwt.2013.04.008
Reis N, Franca AS, Oliveira LS (2013b) Quantitative evaluation of mul-
tiple adulterants in roasted coffee by diffuse reflectance infrared
Fourier transform spectroscopy (DRIFTS) and chemometrics.
Tal ant a 11 5 :56 3–568. https://doi.org/10.1016/j.talanta.2013.06.004
Reis N, Franca AS, Oliveira LS (2016) Concomitant use of Fourier trans-
form infrared attenuated total reflectance spectroscopy and
chemometrics for quantification of multiple adulterants in roasted
and ground coffee. Journal of Spectroscopy 2016. https://doi.org/10.
1155/2016/4974173
Ribeiro JS, Ferreira MMC, Salva TJG (2011) Chemometric models for
the quantitative descriptive sensory analysis of Arabica coffee bev-
erages using near infrared spectroscopy. Talanta 83(5):1352–1358.
https://doi.org/10.1016/j.talanta.2010.11.001
Santos KM, Moura MFV, Azevedo FG, Lima K, Raimundo IM Jr,
Pasquini C (2012) Classification of Brazilian coffee using near-
infrared spectroscopy and multivariate calibration. Anal Lett 45(7):
774–781. https://doi.org/10.1080/00032719.2011.653905
SCA: Specialty Coffee Association of America (2015). http://www.SCA.
org/. Accessed 09 November 2015
Silverstein MR, Webster FX (1998) Identificação Espectrométrica de
Compostos Orgânicos. LTC, Rio de Janeiro
Sunaharum WB, Williams DJ, Smyth HE (2014) Complexity of coffee
flavor: a compositional and sensory perspective. Food Res Int 62:
315–325. https://doi.org/10.1016/j.foodres.2014.02.030
Tolessa K, Rademaker M, De Baets B, Boeckx P (2016) Prediction of
specialty coffee cup quality based on near infrared spectra of green
coffee beans. Talanta 150:367–374. https://doi.org/10.1016/j.
talanta.2015.12.039
Wang J, Ju S, Bittenberden HC, Gautz L, Li QX (2009) Fourier transform
infrared spectroscopy for Kona coffee authentication. Food Sci 74:
358–391. https://doi.org/10.1111/j.1750-3841.2009.01173.x
Wang N, Fu Y, Lim LT (2011) Feasibility study on chemometric discrim-
ination of roasted arabica coffees by solvent extraction and Fourier
transform infrared spectroscopy. J Agric Food Chem 59(7):3220–
3226. https://doi.org/10.1021/jf104980d
Wang N, Linn LT (2012) Fourier transform infrared and physicochemical
analyses of roasted coffee. J Agric Food Chem 60:5446–5453.
https://doi.org/10.1021/jf300348e
Wei F, Tanokura M (2015) Chemical changes in the components of coffee
beans during roasting. In: Preedy VR (ed) Coffee in health and
disease prevention. Elsevier, Barking, pp 83–91
Publisher’sNote Springer Nature remains neutral with regard to jurisdic-
tional claims in published maps and institutional affiliations.
Food Anal. Methods