Food Research International 90 (2016) 133–138

Contents lists available at ScienceDirect

Food Research International

journal homepage: www.elsevier.com/locate/foodres

Differentiation of cocoa nibs from distinct origins using comprehensive

two-dimensional gas chromatography and multivariate analysis
Luciana F. Oliveira a, Soraia C.G.N. Braga a, Fabio Augusto a, Juliana C. Hashimoto b,
Priscilla Efraim b, Ronei J. Poppi a,⁎
a
Institute of Chemistry, University of Campinas, P. O. Box 6194, 13084-971 Campinas, São Paulo, Brazil
b
Department of Food Technology, Faculty of Food Engineering, University of Campinas, P.O. Box 6121, 13083-862 Caempinas, São Paulo, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Headspace solid-phase microextraction (HS-SPME) combined with comprehensive two-dimensional gas chro-
Received 11 July 2016 matography was used for identification of volatiles compounds responsible for differentiation between cocoa
Received in revised form 25 October 2016 nibs from Brazil and Ivory Coast. The unfolded GC × GC with Flame Ionization Detector (FID) chromatograms
Accepted 29 October 2016
were first aligned using the correlation optimization warping (COW) algorithm and normalized. After that, Fisher
Available online 01 November 2016
ratio was calculated for each variable, and a threshold value was chosen to select the variables that best promote
Keywords:
a separation of nibs samples from different sources in a principal component analysis (PCA) model. To identify
Cocoa nibs the relevant compounds for the separation, representative samples of each source were analyzed in the same
Solid phase microextraction conditions by GC × GC with detection by quadrupole mass spectrometry. Finally, the average peak volumes for
Comprehensive two-dimensional gas chroma- each key compound obtained for both classes were compared using a Student t-test and it was possible to identify
tography 15 volatile compounds responsible for differentiation between cocoa nibs from Brazil and Ivory Coast.
Fisher ratio © 2016 Elsevier Ltd. All rights reserved.
Principal component analysis

1. Introduction analytical techniques used for differentiation of cocoa from different

Cocoa nibs, a precursor in making processed chocolate, are obtained
after several preprocess steps, where the cocoa bean is fermented, dried,
roasted and crushed (Payne, Hurst, Miller, Rank, & Stuart, 2010). The
control of these steps is essential to produce cocoa nibs with reliable
characteristics, where the fermentation and roasted processes are pri-
mordial in the quality of the obtained nibs (Frauendorfer & Schieberle,
2008). The volatile compounds present in the cocoa are responsible
for organoleptic properties, such as aroma and flavor. It is known that
compounds such as pyrazines, aldehydes (cocoa aroma, nut), esters
(fruity aroma), and phenolic compounds (astringent property) are di-
rectly related to aroma and they can be used for cocoa and chocolate
characterization (Cambrai et al., 2010).
Depending of the origin of the cocoa nibs, different sensorial and
chemical characteristics will be transferred to the produced chocolate.
Thereby, the knowledge of the volatile compounds present in cocoa
nibs from distinct sources is important to determine the quality of choc-
olate and the veracity of labeling (Saltini, Akkerman, & Frosch, 2013).
The discrimination between cocoa (and derivatives) with different
characteristics has been described in the literature (Cambrai et al.,
2010; Hernandez & Hernandez & Rutledge, 1994; Nightingale,
Cadwallader, & Engeseth, 2012; Yanus et al., 2014). Among the several
⁎ Corresponding author.
E-mail address: ronei@iqm.unicamp.br (R.J. Poppi).
origins, data fusion of near infrared spectroscopy and electronic tongue
was applied for classification of cocoa beans varieties (Teye, Huang,
Takrama, & Haiyang, 2014). More recently, Hori, Kiriyama, and
Tsumura (2016) described the use of liquid chromatography and partial
least squares-discriminant analysis (PLS-DA) in order to distingue cocoa
beans from different growing regions (Dominica, Madagascar, Ghana,
Ivory Coast, Venezuela and Ecuador) and phenolic compounds were in-
dicated as important in this PLS-DA model. Also, gas chromatography
using FID or mass spectrometry detection has been used to distingue dif-
ferent cocoa and derivatives samples (Cambrai et al., 2010; Frauendorfer
& Schieberle, 2008; Rodriguez-Campos, Escalona-Buendía, Orozco-Avila,
Lugo-Cervantes, & Jaramillo-Flores, 2011).
Comprehensive two dimensional gas chromatography presents sev-
eral advantages when compared with traditional gas chromatography,
mainly to promote the increase in the resolution power, since co-eluted
compounds in the first dimension can be separated in the second one
(Marriott, Chin, Maikhunthod, Schmarr, & Bieri, 2012). Moreover
GC × GC chromatograms are structured, facilitating the identification
of unknown compounds (Adahchour, Beens, & Brinkman, 2008; Fitz
et al., 2014; Pierce, Hoggard, Mohler, & Synovec, 2008). For all these rea-
sons GC × GC have been used to analyze a wide range of complex matri-
ces (Dugo et al., 2014; Maikhunthod & Marriott, 2013; Radovic et al.,
2014; Welke, Manfroi, Zanus, Lazzarotto, & Alcaraz Zini, 2013; Zeng,
Li, Hugel, Xu, & Marriott, 2014). However, comprehensive two dimen-
sional gas chromatography of complex samples generates large data

http://dx.doi.org/10.1016/j.foodres.2016.10.047
0963-9969/© 2016 Elsevier Ltd. All rights reserved.
134 L.F. Oliveira et al. / Food Research International 90 (2016) 133–138
sets, mainly if the detection is performed by mass spectrometers, since
an ion fragment dimension comes up in function of the mass spectra ob-
tained for each retention time (Matos, Duarte, & Duarte, 2012). This in-
trinsic fact makes necessary the use of chemometric methods in
GC × GC for data analysis and interpretation.
Several chemometric methods have been successfully applied in
GC × GC data analysis for classification or multivariate calibration
(Gröger & Zimmermann, 2011; Hantao et al., 2013a; Kehimkar et al.,
2014; Vial et al., 2011; Wei et al., 2013; Yang, Hoggard, Lidstrom, &
Synovec, 2013). In order to improve the separation between classes,
variable selection can be an interesting alternative since it turns the
model less complex and easy for interpretation by the fact that unim-
portant variables are removed (Brown, Spiegelman, & Denham, 1991).
Variable selection can be performed in different ways, such as loadings
interpretation in Principal Component Analysis (PCA) models or using
heuristic algorithms as genetic algorithm (Araújo et al., 2001; Jarvis &
Goodacre, 2005).
Fisher ratio is another method that has been applied in feature selec-
tion including bi-dimensional chromatography for various matrices
(Hantao et al., 2013b; Marney et al., 2013; Parsons et al., 2015;
Pinkerton, Parsons, Anderson, & Synovec, 2015). In a simple way, this
methodology can be described as: with the defined classes, a variation
between classes for one variable is calculated and for the same variable
the variation inside the class is also calculated. If the ratio between these
two calculated variations is high, it means that the variable is important
to discriminate the class.
The main purpose of this work was the identification of volatile com-
pounds responsible for differentiation between cocoa nibs from Brazil
and Ivory Coast using headspace solid-phase microextraction (HS-
SPME) combined with comprehensive two-dimensional gas chroma-
tography. Fisher ratio was used for selection of the variables responsible
for separation of nibs samples from the different sources in a principal
component analysis model.
2. Material and methods
2.1. Samples and extraction materials
Cocoa nibs samples fermented, dried and certified from three differ-
ent states in Brazil ( being 7 samples from state of Pará (PA), 8 samples
from state of Bahia (BA) and 7 samples from state of Rondônia (RO))
and Ivory Coast (6 samples) were provided by a local cocoa processing
company. Immediately after receipt the samples, they were packed in
metallized polyethylene bags in temperature of 4 °C until analysis.
The isolation of the volatile fraction of the samples was performed
by HS-SPME, using a procedure previously optimized (Oliveira, Braga,
Filgueiras, Augusto, & Poppi, 2014). Aliquots of (1.000 ± 0.005 g) of
finely ground sample were enclosed in 15 mL vials with 2 mL of NaCl
satured solution. After a 5 min period for sample/headspace equilibra-
tion at 60 °C, a 50/30 μm divinilbenzene/carboxen/polydimethylsilox-
ane (DVB/CAR/PDMS) fiber was exposed to the sample headspace for
50 min; the fiber was then immediately exposed for 5 min to the injec-
tor of the GC × GC systems. Before the extractions the fiber was condi-
tioned for 2 h at 250 °C according manufacturer's instructions.
2.2. GC × GC-FID and GC × GC-QMS
GC × GC-FID analysis were carried out on a prototype based on an
Agilent 6890 GC fitted with a lab-made 4-jet cryogenic modulator,
already described on the literature (Pedroso, Ferreira, Hantao, Bogusz,
& Augusto, 2011). The column set consisted on a 30 m ×
0.25 mm × 0.25 μm HP-5 column (Agilent Technologies, Wilmington,
DE) connected to a 0.80 m × 0.1 mm × 0.1 μm Solgel Wax column
(SGE Analytical Science, Ringwood - Victoria, Australia). The modula-
tion period was set to 6.0 s. The split-splitless injector was operated
on splitless mode at 260 °C. The column oven temperature was
programmed as follows: 40 °C to 110 °C at 3 °C min−1 and 110 °C to
240 °C at 10 °C min−1 (staying during 4 min at this temperature
− 240 °C). High purity (99.9999%) hydrogen at 0.6 mL min− 1 was
used as carrier gas and the acquisition rate of FID detector was 100 Hz.
The same conditions were used in GC × GC–MS analysis, using a pro-
totype built over a Shimadzu QP2010 + fast GC-QMS gas chromato-
graph (Hantao et al., 2013b). In this case the interface temperature
was 260 °C and photomultiplier power was programmed to 0.8 kV up
to 10 min of the chromatographic run, and then increased to 0.9 kV
until final. The scanned mass range was set from m/z = 40 D to 340 D,
resulting on a data collection frequency of 25 spectra s−1. Also, for
the calculation of 1st dimension linear temperature programming re-
tention indexes (LTPRI) of relevant peaks, analysis of samples spiked
with n-alkane mixture (C8 to C20) were also performed. Processing of
the chromatographic data was performed using the GCImage software
(Zoex Corp., Houston – TX, USA), and identification of the peaks on
the GC × GC-QMS chromatograms was performed after search on NIST
2010 mass spectra library.
2.3. Alignment of signal on GC × GC chromatograms
The unfolded chromatograms were aligned in the time axis using the
Correlation Optimization Warping (COW) algorithm. First, the chro-
matograms were split in 10 sections order to minimize the number of
points in the optimization step (best number of slack size, window
and the reference chromatogram) as described for Skov, van den Berg,
Tomasi, and Bro (2006)). After that, the COW algorithm was applied
separately for each section of the chromatograms; the aligned sections
were merged resulting complete chromatograms which were used to
perform Fisher Ratio analysis.
2.4. Fisher ratio and PCA analysis
The sample set was divided in two groups corresponding to Brazilian
and Ivorian samples. Fisher ratios were calculated for each variable (i.e.,
discrete signal points on the 1tR × 2tR plane) according to Eq. (3), using
the unfolded GC × GC-FID chromatograms. The variation between
the two classes, inside the class, and the ratio is calculated using
Eqs. (1)–(3), respectively (Pierce et al., 2005).
∑ðxi −xÞ2 ni
σ 2cl ¼ ð1Þ
ðk−1Þ
where ni is the number of variables in the ith class, xi is the mean for the
ith class, x is the global mean, and k is the number of classes.

2   
∑ ∑ xij –x − ∑ðxi −xÞ2 ni
σ 2err ¼ ð2Þ
N−k
where xij is the ith variable of the jth class and N is the total number of
samples.
σ 2cl
f ratio ¼ ð3Þ
σ 2err
These Fisher ratios consist on a vector of scalars one for each vari-
able: wherever a value is higher than a defined threshold, it is supposed
that there is a peak relevant to the differentiation between the classes
(Brazilian and Ivorian samples) on the corresponding position on the
chromatographic 1tR × 2tR plane – which was then located on the
GC × GC-QMS chromatograms and identified.
In order to visualize the class separation after the variable selection
with Fisher ratios a Principal Component Analysis with normalized
data was performed using Matlab routines. Assessments of scores
plots were used to define the best threshold and consequently the
most important variables selected by Fisher ratios. All data analysis
 L.F. Oliveira et al. / Food Research International 90 (2016) 133–138 135

GCxGC-FID
chromatogramns alignment
chromatograms unfolded chromatograms with COW algorithm

K(2tr) samples samples

samples
J(1tr) K(2tr)*J(1tr) K(2tr)*J(1tr)

Fisher ratios foreach variable inunfolded

chromatograms (samples from two classes)
folded back
Fisher ratios
or
samples (high values -
unfolded Fisher ratios
K(2tr)*J(1tr) important
variables)

Fig. 1. General representation of alignment and Fisher ratio analysis.

was performed in Matlab environment version R2014a (MathWorks, input data and only using the variables previously screened as poten-
Natick – MA, USA) using homemade routines and PLS_Toolbox (version tially relevant to the class differentiation (i.e., discarding sections of
7.3.1 – Eigenvector, WA, USA). Fig. 1 shows a general representation of the chromatograms which were found not to contain relevant informa-
the data processing described on Sections 2.3 and 2.4. tion for the discrimination of the classes). This variable selection was
Finally, to ensure that the Fisher ratio procedure selected correctly performed assessing Fisher ratios for each variable in the
the compounds responsible for class separation, the volume of the chro- chromatograms.
matographic peak for each compound was estimated for all original Fig. 2a shows a plot of the first two PCA scores obtained when raw
chromatograms and the average peak volumes for each key compound aligned chromatograms are used as input data: no class differentiation
obtained for both classes were compared using a Student t-test. is discernible. As for PCA results using data sets after removal of non-
significant variables through Fisher ratio analyses, the success on
obtaining class differentiation depended on the threshold selected for
3. Results and discussion fratio (Eq. (3)), these threshold values were chosen by analyzing the
scores plots in PCA model and considering the values of Fisher ratios.
The overall experimental approach here was divided on two main Separate PCA modelling using data sets obtained using different thresh-
steps: initially, all samples were analyzed using GC × GC-FID, and after old values - 3 (6137 variables selected), 7 (1778 variables selected) and
alignment and evaluation using Fisher ratio and PCA analysis, 11 (959 variables selected) - were performed (Fig. 2b–d). On these fig-
GC × GC-QMS chromatograms for selected samples were obtained and ures it can be seen that differentiation of samples according to their clas-
the relevant sections pointed out by the later processing inspected to ses becomes clearer when the rigor on preliminary variable selection is
identify key compounds that differentiate the sample classes (Brazilian increased using larger threshold values for fratio. For the model obtained
and Ivorian cocoa nibs). Visual comparison of chromatograms obtained with a threshold of 3, 83% of the total variance is explained but there is
for both classes (see Fig. 1S in Supplementary material) showed that it is no clear separation of classes. However, when using 11 as threshold, the
possible to observe differences in intensities of some peaks between the data sets corresponding to Brazilian and Ivorian samples are clearly
samples from Brazil and Ivory Coast; due to the complexity of the chro- grouped, and 94% of the total variance is explained by the resulting
matograms, multivariate analysis is necessary to find out which parts of PCA model.
the chromatograms and which compounds can be used to discern the Fig. 3 shows a plot of the Fisher ratios calculated from the raw
classes. aligned chromatograms, with indication of the threshold value of 11 –
Before application of any multivariate analysis tool to GC × GC data it determined as the most adequate for PCA modelling (each variable cor-
is mandatory to perform signal alignment between the chromatograms respond to an interval of 10,42 ms on the 1st dimension time axis on the
in the temporal domain. This procedure is fundamental to ensure that chromatograms). This plot containing 959 signal values with Fisher
the variables (discrete signal points on the 1tR × 2tR plane) represent ratio ≥ 11 can be folded and used to generate a figure where the position
the same retention times in all input data used in the multivariate anal- of chromatographic peaks relevant to class differentiation can be
ysis. To perform this task, the COW (correlation optimization warping)
algorithm was applied to the unfolded chromatograms. This algorithm
works dividing the chromatograms in segments with different length
and the move boundaries between segments are defined with certain Table 1
quantity of points (called slack size in COW algorithm). A chromato- The optimized chromatogram used as reference, segment length and slack size values for
each part of the chromatograms.
gram from the subset being aligned is selected as reference for the align-
ment, and the remainders are adjusted according to this base Subset intervals (s) Reference chromatogram # Segment length Slack size
chromatogram. These parameters were optimized as described for 0–230,21 3 175 9
Skov et al. (2006). The slack size, the segment length and the chromato- 230,22–555,79 30 400 14
gram used as reference in the alignment of each part of the data set are 559,80–705,00 1 400 11
705,01–941,98 18 325 13
shown in Table 1.
941,99–1042,83 1 100 14
After the procedure, the chromatograms were properly aligned in 1042,84–1260,00 1 101 14
the time domain (q.v. Fig. 2S in Supplementary material) and ready 1260,01–1485,94 15 400 14
for multivariate analysis. PCA was performed to differentiate cocoa sam- 1485,95–1680,00 26 399 15
ples from Brazil and Ivory Coast, using two different approaches: using 1680,01–1886,25 18 401 15
1886,26–2100,00 29 100 10
the complete aligned chromatograms (matrixes signal × 1tR × 2tR) as
136 L.F. Oliveira et al. / Food Research International 90 (2016) 133–138

Samples/Scores Plot Samples/Scores Plot

6 1

4 0.5

Scores on PC 2 (17.94%)

Scores on PC 2 (19.16%)
2 0

0 -0.5

-2 -1

-4 -1.5

-6 -2
-6 -4 -2 0 2 4 6 -2 -1.5 -1 -0.5 0 0.5 1 1.5
Scores on PC 1 (31.14%) Scores on PC 1 (64.09%)

(a) (b)
Samples/Scores Plot Samples/Scores Plot
0.6 0.6

0.4 0.5
Scores on PC 2 (6.26%)

Scores on PC 2 (3.65%)
0.4
0.2

0.3
0

0.2

-0.2
0.1

-0.4
0

-0.6 -0.1

-0.8 -0.2
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1

Scores on PC 1 (87.92%) Scores on PC 1 (90.78%)

(c) (d)
Fig. 2. PCA scores plot (a) threshold equal to 0 and all the variables included in PCA model; (b) threshold equal to 3 and 6137 variables included in PCA model, (c) threshold equal to 7 and
1778 variables included in PCA model and (d) threshold equal to 11 and 959 variables included in the PCA model. ○ samples from Brazil and ● samples from Ivory Coast.

visually located on the 1tR × 2tR plane. Superposition of this plot to the To further assess the correlation between these compounds and the
chromatograms (Fig. 4) shown that 20 plots corresponded to chromato- differentiation between the classes of samples, the volume of the select-
graphic peaks, which were then assigned as for separation between Bra- ed chromatographic peaks (parameter corresponding to the peak area
zilian and Ivorian samples according to the composition of their in conventional one-dimensional chromatography) was estimated for
headspaces. all chromatograms and the average peak volumes for each key
To identify the compounds responsible for the class separation, rep- compound obtained for both classes were compared using a simple
resentative samples from Brazil and Ivory Coast were further analyzed null hypothesis Student t-test. For three compounds (limonene, 2-iso-
by GC × GC-QMS, using as much as possible the same conditions propyl-5-metil-hex-2-enal and dihidro-2(3H)-furanone) the peak vol-
adopted on the previous experiments. The tentative identification the umes – and, therefore, their average concentrations in samples – were
relevant peaks was performed matching their mass spectra with the not found to be significantly different between classes with 95% confi-
NIST08 library (minimum similarity of 75%); these identifications dence. The concentrations of remaining 15 compounds were deter-
were confirmed by comparison of the corresponding LTPRI with those mined to be significantly different between Brazilian and Ivorian
of standards, when available, or by assessment of the structuration on cocoa samples. To simplify the visualization of the results, the average
the GC × GC chromatograms. The identified compounds are described volume for each compound was normalized defining the larger value
in Table 2; 2 out of the 20 compounds could not be identified. of the pair as 1.0 (Fig. 5).

Fisher Ratios
80

60
Fisher ratios

40

20

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Variables

Fig. 3. Calculated Fisher ratio values for all variables in the chromatogramns. The horizontal line corresponds to the best threshold value.
 L.F. Oliveira et al. / Food Research International 90 (2016) 133–138 137

600

550

500

450

points in 2nd dimension

400

350

300

250

200

150

100

50

50 100 150 200 250 300 350

points in1st dimension

Fig. 4. Representative Ivory Coast chromatogram in black and in red the spots responsible for samples separation (threshold in Fisher ratio above to 11). (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2
Compounds responsible for differentiation between cocoa nibs from Brazil and Ivory Coast and respective retention time in first and second dimension.

Compound Rt1 (min) Rt2 (s) Compound Rt1 (min) Rt2 (s)

1 Pentanal 4.00 1.27 10 Benzaldehyde 12.6 2.22

2 (2E)-2-pentenal 5.30 2.11 11 2-Pentyl-furan 13.8 1.91
3 Hexanal 6.60 1.82 12 2-Ethyl-5-methyl-pyrazine 14.1 3.47
4 Butyl acetate 6.70 1.54 13 Trimethyl-pyrazine 14.4 3.67
5 5-Methyl-5-hexen-2-one 6.80 1.34 14 Limonene 15.5 1.57
6 Furfural 7.90 5.18 15 2,3,3-Trimethyl-4-nonene 16.5 2.06
7 Nonane 10.4 1.04 16 2-Methyl decane 16.8 1.67
8 (Z)-2-heptenal 10.6 2.82 17 1-Methyl-2-oxopropyl ester 17.1 2.05
9 Dihydro-2(3H)-furanone 10.9 4.81 18 2-Isopropyl-5-methylhex-2-enal 18.2 2.23

Some of these compounds have been already found to be relevant on 1991). Furthermore, benzaldehyde was described for Cambrai et al.,
the characterization of flavor of cocoa products and related materials. 2010 as typical from chocolates made with African cocoa.
For example, esters such as butyl acetate (#4 on Table 2) are responsible
for fruity notes on chocolate and cocoa flavor; aldehydes such as furfu- 4. Conclusions
ral, benzaldehyde and hexanal (#6, #10 and #3) are listed as character-
istics flavors from hazelnut and almond (Counet, Ouwerx, Rosoux, & The Fisher ratios proved an extremely simple and powerful way to
Collin, 2004; Hernandez & Hernandez & Rutledge, 1994; Ziegleder, isolate, in complex data sets and higher order, the chromatographic

1.2
Ivory Coast
Normalized average peak volumes

Brazil

0.8

0.6

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Compounds

Fig. 5. Normalized average peak volumes for compounds found to be relevant to class differentiation. The larger peak volume for each compound was defined as 1.00; compound
identification according to peak numbers on Table 2.
138 L.F. Oliveira et al. / Food Research International 90 (2016) 133–138
sections containing relevant information to the class separation. The ef-
fectiveness of this strategy could be proven by the successful use of PCA
after the selection of variables based on the Fisher ratios. This combina-
tion of two chemometric tools relatively unsophisticated can prove to
be a powerful strategy in the GC × GC data evaluation,
Moreover, the use of Fisher ratio in conjunction with principal com-
ponent analysis applied to HS-SPME GC × GC with FID detection make
possible to find 20 compounds responsible for differentiation between
cocoa nibs from Brazil and Ivory Coast. Using a quadrupole mass detec-
tor and the NIST library, 18 compounds could be identified and from the
average values of these compounds, and applying a t-test, was proved
that 15 compounds have significant different contents in samples
from Brazil and Ivory Coast.
Acknowledgements
The authors thank Fundação de Amparo à Pesquisa do Estado de
São Paulo – FAPESP (proc. 2012/14695-8), Conselho Nacional de
Desenvolvimento Científico e Tecnológico - CNPq (proc. 307838/
2013-7) for financial support.
Appendix A. Supplementary data
Supplementary data to this article can be found online at doi:10.
1016/j.foodres.2016.10.047.
References
Adahchour, M., Beens, J., & Brinkman, U. A. T. (2008). Recent developments in the appli-
cation of comprehensive two-dimensional gas chromatography. Journal of
Chromatography A, 1186(1–2), 67–108.
Araújo, M. C. U., Saldanha, T. C. B., Galvão, R. K. H., Yoneyama, T., Chame, H. C., & Visani, V.
(2001). The successive projections algorithm for variable selection in spectroscopic
multicomponent analysis. Chemometrics and Intelligent Laboratory Systems, 57(2),
65–73.
Brown, P. J., Spiegelman, C. H., & Denham, M. C. (1991). Chemometrics and spectral fre-
quency selection. Philosophical Transactions of the Royal Society A: Mathematical,
Physical and Engineering Sciences, 337(1647), 311–322.
Cambrai, A., Marcic, C., Morville, S., Sae Houer, P., Bindler, F., & Marchioni, E. (2010). Dif-
ferentiation of chocolates according to the Cocoa's geographical origin using
chemometrics. Journal of Agricultural and Food Chemistry, 58(3), 1478–1483.
Counet, C., Ouwerx, C., Rosoux, D., & Collin, S. (2004). Relationship between procyanidin
and flavor contents of cocoa liquors from different origins. Journal of Agricultural and
Food Chemistry, 52(20), 6243–6249.
Dugo, G., Franchina, F. A., Scandinaro, M. R., Bonaccorsi, I., Cicero, N., Tranchida, P. Q., &
Mondello, L. (2014). Elucidation of the volatile composition of Marsala wines by
using comprehensive two-dimensional gas chromatography. Food Chemistry, 142,
262–268.
Fitz, B. D., Reaser, B. C., Pinkerton, D. K., Hoggard, J. C., Skogerboe, K. J., & Synovec, R. E.
(2014). Enhancing gas chromatography − time of flight mass spectrometry data
analysis using two-dimensional mass channel cluster plots. Analytical Chemistry, 86,
3973–3979.
Frauendorfer, F., & Schieberle, P. (2008). Changes in key aroma compounds of Criollo
cocoa beans during roasting. Journal of Agricultural and Food Chemistry, 56(21),
10244–10251.
Gröger, T., & Zimmermann, R. (2011). Application of parallel computing to speed up
chemometrics for GC × GC–TOFMS based metabolic fingerprinting. Talanta, 83(4),
1289–1294.
Hantao, L. W., Toledo, B. R., de Lima Ribeiro, F. A., Pizetta, M., Pierozzi, C. G., Furtado, E. L., &
Augusto, F. (2013a). Comprehensive two-dimensional gas chromatography com-
bined to multivariate data analysis for detection of disease-resistant clones of euca-
lyptus. Talanta, 116, 1079–1084.
Hantao, L. W., Aleme, H. G., Passador, M. M., Furtado, E. L., de Lima Ribeiro, F. A., Poppi, R. J.,
& Augusto, F. (2013b). Determination of disease biomarkers in eucalyptus by com-
prehensive two-dimensional gas chromatography and multivariate data analysis.
Journal of Chromatography A, 1279, 86–91.
Hernandez, C. V., & Rutledge, D. N. (1994). Multivariate statistical analysis of gas chro-
matograms to differentiate cocoa masses by geographical origin and roasting condi-
tions. Analyst, 119(6), 1171–1176.
Hori, K., Kiriyama, T., & Tsumura, K. (2016). A liquid chromatography time-of-flight mass
spectrometry-based metabolomics approach for the discrimination of cocoa beans
from different growing regions. Food Analytical Methods, 9(3), 738–743.
Jarvis, R. M., & Goodacre, R. (2005). Genetic algorithm optimization for pre-processing
and variable selection of spectroscopic data. Bioinformatics, 21(7), 860–868.
Kehimkar, B., Hoggard, J. C., Marney, L. C., Billingsley, M. C., Fraga, C. G., Bruno, T. J., &
Synovec, R. E. (2014). Correlation of rocket propulsion fuel properties with chemical
composition using comprehensive two-dimensional gas chromatography with time-
of-flight mass spectrometry followed by partial least squares regression analysis.
Journal of Chromatography A, 1327, 132–140.
Maikhunthod, B., & Marriott, P. J. (2013). Aroma-impact compounds in dried spice as a
quality index using solid phase microextraction with olfactometry and comprehen-
sive two-dimensional gas chromatography. Food Chemistry, 141(4), 4324–4332.
Marney, L. C., Christopher Siegler, W., Parsons, B. A., Hoggard, J. C., Wright, B. W., &
Synovec, R. E. (2013). Tile-based Fisher-ratio software for improved feature selection
analysis of comprehensive two-dimensional gas chromatography–time-of-flight
mass spectrometry data. Talanta, 115, 887–895.
Marriott, P. J., Chin, S. -T., Maikhunthod, B., Schmarr, H. -G., & Bieri, S. (2012). Multidimen-
sional gas chromatography. TrAC Trends in Analytical Chemistry, 34, 1–21.
Matos, J. T. V., Duarte, R. M. B. O., & Duarte, A. C. (2012). Trends in data processing of com-
prehensive two-dimensional chromatography: State of the art. Journal of
Chromatography B, 910, 31–45.
Nightingale, L. M., Cadwallader, K. R., & Engeseth, N. J. (2012). Changes in dark chocolate
volatiles during storage. Journal of Agricultural and Food Chemistry, 60(18),
4500–4507.
Oliveira, L. F., Braga, S. C. G. N., Filgueiras, P. R., Augusto, F., & Poppi, R. J. (2014). Assess-
ment of robustness on analysis using headspace solid-phase microextraction and
comprehensive two-dimensional gas chromatography through experimental de-
signs. Talanta, 129, 303–308.
Parsons, B. A., Marney, L. C., Siegler, W. C., Hoggard, J. C., Wright, B. W., & Synovec, R. E.
(2015). Tile-based Fisher ratio analysis of comprehensive two-dimensional gas chro-
matography time-of-flight mass spectrometry (GC × GC–TOFMS) data using a null
distribution approach. Analytical Chemistry, 87(7), 3812–3819.
Payne, M. J., Hurst, W. J., Miller, K. B., Rank, C., & Stuart, D. A. (2010). Impact of fermenta-
tion, drying, roasting, and dutch processing on epicatechin and catechin content of
cacao beans and cocoa ingredients. Journal of Agricultural and Food Chemistry,
58(19), 10518–10527.
Pedroso, M. P., Ferreira, E. C., Hantao, L. W., Bogusz, S., & Augusto, F. (2011). Identification
of volatiles from pineapple (Ananas comosus L.) pulp by comprehensive two-dimen-
sional gas chromatography and gas chromatography/mass spectrometry. Journal of
Separation Science, 34(13), 1547–1554.
Pierce, K. M., Hoggard, J. C., Hope, J. L., Rainey, P. M., Hoofnagle, A. N., Jack, R. M., ...
Synovec, R. E. (2005). Fisher ratio method applied to third-order separation data to
identify significant chemical components of metabolite extracts. Analytical
Chemistry, 78(11), 218–225.
Pierce, K. M., Hoggard, J. C., Mohler, R. E., & Synovec, R. E. (2008). Recent advancements in
comprehensive two-dimensional separations with chemometrics. Journal of
Chromatography A, 1184(1–2), 341–352.
Pinkerton, D. K., Parsons, B. A., Anderson, T. J., & Synovec, R. E. (2015). Trilinearity devia-
tion ratio: A new metric for chemometric analysis of comprehensive two-dimension-
al gas chromatography time-of-flight mass spectrometry data. Analytica Chimica Acta,
871, 66–76.
Radovic, J. R., Thomas, K. V., Parastar, H., Sergi, D., Tauler, R., & Bayona, J. M. (2014).
Chemometrics-assisted effect-directed analysis of crude and refined oil using com-
prehensive two-dimensional gas chromatography − time-of-flight mass spectrome-
try. Environmental Science and Technology, 48, 3074–3083.
Rodriguez-Campos, J., Escalona-Buendía, H. B., Orozco-Avila, I., Lugo-Cervantes, E., &
Jaramillo-Flores, M. E. (2011). Dynamics of volatile and non-volatile compounds in
cocoa (Theobroma cacao L.) during fermentation and drying processes using principal
components analysis. Food Research International, 44(1), 250–258.
Saltini, R., Akkerman, R., & Frosch, S. (2013). Optimizing chocolate production through
traceability: A review of the influence of farming practices on cocoa bean quality.
Food Control, 29, 167–187.
Skov, T., van den Berg, F., Tomasi, G., & Bro, R. (2006). Automated alignment of chromato-
graphic data. Journal of Chemometrics, 20, 484–497.
Teye, E., Huang, X., Takrama, J., & Haiyang, G. (2014). Integrating NIR spectroscopy and
electronic tongue together with chemometric analysis for accurate classification of
cocoa bean varieties. Journal of Food Process Engineering, 37(6), 560–566.
Vial, J., Pezous, B., Thiébaut, D., Sassiat, P., Teillet, B., Cahours, X., & Rivals, I. (2011). The
discriminant pixel approach: A new tool for the rational interpretation of GC × GC–
MS chromatograms. Talanta, 83(4), 1295–1301.
Wei, X., Shi, X., Koo, I., Kim, S., Schmidt, R. H., Arteel, G. E., ... Zhang, X. (2013). MetPP: A
computational platform for comprehensive two-dimensional gas chromatography
time-of-flight mass spectrometry-based metabolomics. Bioinformatics, 29(14),
1786–1792.
Welke, J. E., Manfroi, V., Zanus, M., Lazzarotto, M., & Alcaraz Zini, C. (2013). Differentiation
of wines according to grape variety using multivariate analysis of comprehensive
two-dimensional gas chromatography with time-of-flight mass spectrometric detec-
tion data. Food Chemistry, 141(4), 3897–3905.
Yang, S., Hoggard, J. C., Lidstrom, M. E., & Synovec, R. E. (2013). Gas chromatography and
comprehensive two-dimensional gas chromatography hyphenated with mass spec-
trometry for targeted and nontargeted metabolomics. Metabolomics in practice
(pp. 69–92). Weinheim, Germany: Wiley-VCH Verlag GmbH & Co. KGaA.
Yanus, R. L., Sela, H., Borojovich, E. J. C., Zakon, Y., Saphier, M., Nikolski, A., ... Karpas, Z.
(2014). Trace elements in cocoa solids and chocolate: An ICPMS study. Talanta, 119,
1–4.
Zeng, Z., Li, J., Hugel, H. M., Xu, G., & Marriott, P. J. (2014). Interpretation of comprehensive
two-dimensional gas chromatography data using advanced chemometrics. TrAC
Trends in Analytical Chemistry, 53, 150–166.
Ziegleder, G. (1991). Composition of flavor extracts of raw and roasted cocoas. European
Food Research and Technology, 192(6), 521–525.