mistry & A

he

pp
Agatonovic-Kustrin et al., Mod Chem appl 2013, 1:4
Modern C Modern Chemistry & Applications
lications
DOI: 10.4172/2329-6798.1000110
ISSN: 2329-6798

Research Article Open Access

The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial

Neural Networks (ANNs) to Assess Wine Quality
Snezana Agatonovic-Kustrin1*, David W. Morton1 and Ahmad Pauzi Md. Yusof2
1
School of Pharmacy and Applied Science, La Trobe University, Australia
2
Physiology Department, Medical School, Universiti Teknologi Mara, Selangor, Malaysia

Abstract
The aim of this study was to develop a simple method to assess wine quality from its Fourier Transform Infrared
Spectroscopy (FTIR) spectrum with minimal or no sample preparation. FTIR spectral data of selected wine samples,
grape variety, wine barrel type, wine type and production year were correlated with total phenolic content, total and
volatile acidity and alcohol content using Artificial Neural Networks (ANNs). A total of 20 (2 whites and 18 reds) different
wines used in this study came from three different states across Australia; New South Wales, Victoria and South
Australia.
FTIR spectroscopy proved to be a promising technique that provides a rapid and accurate method in the quality
assessment of wine. A plot of the values predicted by the validated ANN models showed excellent correlation with the
experimentally measured values for acetic acid concentration, alcohol content, total phenols, and total acidity (r=0.898-
0.942).

Keywords: Fourier transform infrared spectroscopy; Artificial neural grape and wine quality is important to the wine industry. It is desirable
network; Wine quality assessment; Polyphenolic content that these methods require minimal sample preparation and are able
to produce rapid results, preferably providing information on multiple
Introduction parameters simultaneously. The aim of this study was to develop a simple
It is known that moderate consumption of red wine and fruit juices and rapid method based on Fourier Transform Infrared Spectroscopy
can reduce the risk for cardiovascular disease [1,2]. The protective effects (FTIR) spectroscopy combined with Artificial Neural Network data
of wine have been attributed to polyphenols that are efficient scavengers modeling to assess multiple quality indicators of wine samples. Using
of free radicals and breakers of lipid peroxidative chain reactions [3]. non-destructive FTIR spectroscopy with a horizontal Attenuated Total
One of the most fascinating observation is the ‘French paradox’, a term Reflectance (ATR) accessory enables wine spectra to be obtained using
first coined by Dr. Serge Renauld, in the 1980s [4,5] that refers to the fact relatively small volumes (around 0.5 to 1 mL) of wine with minimal
that French population have an incredible low coronary heart disease sample preparation and reagent consumption. Mid-IR spectrometry
death rates despite high intake of dietary cholesterol and saturated has been previously used in the analysis of a number of foods, including
fat [5,6]. Numerous studies have documented the health benefits of wine. The typical analytes that have been measured using this technique
red wine consumption, including anti-oxidative, anti-carcinogenic, are ethanol, pH, organic acids, sugars and glycerol [17,18]. However,
anti-inflammatory, anti-cardiovascular and antibacterial properties only a few studies have applied FTIR to analyse selected polyphenolic
[7,8]. Grapes and wines are rich in a large number of polyphenolic compounds such as tannins [19] and anthocyanins [20] and total
compounds, belonging to non-flavonoids, flavonoids and phenolic- antioxidant capacity [21]. The use of Artificial Neural Networks
protein-polysaccharide complexes, which possess high antioxidant (ANNs), as a non-linear statistical data modelling tool was chosen to
activity and are believed to reduce and prevent oxidative stress related correlate FTIR spectra of selected wine samples, grape variety, wine
diseases [9,10], as well as several organic acids, such as tartaric and malic barrel (oak type), wine type (red or white) and production year with
acids, which have antimicrobial effects, especially at the low pH of wine total phenolic content, total and volatile acidity, and alcohol content.
[11,12]. The amount of these potentially beneficial compounds present The aim of this modelling was to: (1) successfully predict the overall
in red wine usually varies depending on the variety of grapes and the polyphenolic content of a wine sample from its FTIR spectra; and (2)
vinification process used [13]. The major polyphenolic compounds to determine how parameters such as alcohol content, pH etc. affect
found in red wine contribute significantly to the major organoleptic the overall polyphenolic content in red wine. This type of information
qualities such as mouth feel, taste and colour, and therefore play an should enable winemakers to better optimise the concentration of
important role in the overall quality of the wine. It is believed that polyphenolic compounds during the winemaking process.
wine quality variations are related to both the origin (structure related)
and quantity (concentration factor) of the polyphenols present in red
wine [14]. Polyphenolics in wine are responsible for varietal and flavor *Corresponding author: Snezana Agatonovic- Kustrin, Associate Professor
characteristics of red wines such as color and tannin characteristics [15]. of Pharmacy, School of Pharmacy and Applied Science, Faculty of Science,
Technology and Engineering, La Trobe University, Bendigo, Post Box 199, Bendigo
Phenolics also affect the sensory characteristics of wines, contributing 3552, Australia, Tel: +61-354-447-360; Fax: +61-354-447-878; E-mail: s.kustrin@
to bitterness and astringency. Astringency and bitterness are produced latrobe.edu.au
primarily by flavonoids that are extracted from the skins and seeds of Received August 13, 2013; Accepted September 14, 2013; Published September
grapes. The types and amounts of different phenolic compounds present 23, 2013
have been used as broad indicators of wine quality and good correlation
Citation: Agatonovic-Kustrin S, Morton DW, Yusof APM (2013) The Use of Fourier
has been obtained between several aspects of the phenolic content and Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
assessed quality of red wine [16]. However, it would be useful to be Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110
able to perform a single measurement that is representative of a wine’s Copyright: © 2013 Agatonovic-Kustrin, et al. This is an open-access article
composition universally correlates with perceived wine quality [17]. distributed under the terms of the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided
The development of more effective and efficient methods to assess the original author and source are credited.

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 2 of 8

Wine sample State/ Region Type of Wine Varietal Composition Year Alcohol Content Type of Oak
1 VIC Red Cabernet 2008 14.5% French
2 VIC Red Cabernet 2006 14% French
Cab. Sauvignon 31%
Merlot 30%
3 VIC Red 2008 15.6%
Cab. Franc 24%
Malbec 15%
Shiraz
4 VIC Red 2007 15% French
Cabernet
5 VIC Red Shiraz 2008 15% French
6 NSW Red Shiraz 2008 13.0% Unknown
7 VIC White Semillon 2009 14% French
Cab. Sauvignon 60%
8 VIC Red Cab. Franc 20% 2008 13.3% French
Merlot 20%
9 VIC White Riesling 2008 13% Tank
10 VIC Red Shiraz 2008 14% American
French
11 SA Red Shiraz 2009 14%
American
Shiraz 96%
12 SA Red 2008 14% American
Petit Verdot 4%

13 VIC Red Shiraz 2010 13.5% French

Shiraz
14 VIC/ SA Red 2009 13.5% Unknown
Merlot
Cabernet
15 VIC Red 2009
Merlot
16 VIC Red Shiraz 2010 15% American
17 VIC Red Shiraz 2001 15% American
18 VIC Red Shiraz 2008 15.5% American
Shiraz 87%
19 VIC Red Malbec 7% 2008 14% American
Cab. Sauvignon 6%
20 VIC Red Cabernet Sauvignon 2008 15% American
Table 1: Wine samples collected from different wineries with the information found on the label.

Materials and Methods were first diluted with distilled water (4.00 mL). An aliquot (0.200 mL)
of the diluted wine sample and 1.00 mL of Folin-Ciocalteu reagent
Wine samples (Sigma Chemicals Co, St. Louis, MO, USA), were added to a 20.00 mL
A total of 20 different wines (2 whites and 18 reds) used in this study volumetric flask. Exactly after 1 minute, 4.00 mL of sodium carbonate
came from three different states across Australia; New South Wales, (20% w/v) (Merck, Vic, Australia) was added and the solution was
Victoria and South Australia. The wines selected varied considerably in made to a total volume of 20.00 mL using distilled water. Finally, the
regards to growing region, varietal composition, year, alcohol content mixture was allowed to stand at room temperature in the dark for 30
and type of oak used (Table 1). min after which time the absorbance of the solution was measured at
750 nm. The total polyphenolic concentration was calculated from a
Total acidity calibration curve using Gallic acid (Aldrich Chemical Co (WI, USA)
as a standard (1-6 mg/L). Polyphenolic concentration was expressed in
Total acidity (free acid) or titratable acidity of the wine samples,
grams of Gallic acid equivalents per Litre (GAE g/L) and was an average
expressed the concentration of tartaric acid (g/L), was determined
of three measurements.
titrimetrically using a standard 0.05M NaOH solution with
phenolphthalein indicator. The color of the end point was a stable Apparatus
gray for red wines and a faint pink color for white wines. In order to
overcome the problem associated of the intense color of the red wines, UV-Vis spectra were collected using a UV-Vis single beam
samples were diluted by 1 in 75 using distilled water prior to the spectrometer (Mini 1240, Shimadzu). The IR spectra of the wine
titration. All pH and total acidity determinations were replicated twice. samples were examined over the range of 400-4000 cm-1 using a Bruker
Equinox 55 FT-IR spectrometer equipped with a horizontal Attenuated
Volatile acidity Total Reflectance (ATR) device with a diamond crystal. Spectra were
The volatile acidity, expressed as the amount of acetic acid (g/L) recorded using OPUS software (Bruker Optik, Germany) by averaging
was determined spectrophotometrically using an enzymatic analysis 100 scans for each spectrum with resolutions of 2 cm-1. Background
kit (Vint essential Laboratories, Dromana, Australia) with absorbance spectra were obtained and subtracted from each sample IR spectra.
measurements made at a wavelength of 340 nm. Small quantities of untreated wine samples were smeared directly onto
the ATR diamond crystal using a disposable pipette. In order to avoid
Total phenolic content the strong interference due to the presence of water and alcohol in each
wine sample, a 12% (v/v) ethanol spectrum was subtracted from each
The total phenolic concentration was determined according to
the Folin-Ciocalteu colorimetric method. Samples of wine (1.00 mL) wine spectrum. Statistical Neural Networks 4.0 F (Stat Soft Inc., Tulsa,

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 3 of 8

OK, USA) was used to model the spectral data and develop predictive superiority in network performance. In contrast to linear statistical
ANN models. techniques, there is no known method for the automatic determination
of an optimal network structure to fit a specific dataset [24]. As a result,
Pre-processing the FTIR data training algorithms are run a number of times through automated
The FTIR spectra were sampled between 400-4000 cm-1 and network searches so that the best networks could be selected.
obtained spectra were smoothed to reduce the noise and improve
Selection of input variables
signal-to noise ratio using the Moving Average method by counting the
average of 20 data sets in every tenth wavelength record [22]. Significant Sensitivity analysis was used for feature selection to identify the most
signal noise in the spectra would degrade the signal-to-noise ratio and important inputs that are directly correlated to total phenolic content,
resolution of the spectra and hence reduce the accuracy and precision total and volatile acidity, pH and alcohol content by determining
of a model. The resulting 185 spectral intensities for each wine sample, how sensitive a model is to changes in the input values. Sensitivity is
together with variety of grape, wine barrel (oak type), and production defined as the ratio of error of a retrained optimum model, which does
year were used as ANN inputs and experimentally measured total not contain the information of a specific molecular descriptor, to the
phenolic content, total acidity, volatile acidity (acetic acid) and pH, error of the optimum model that includes the information from the
were used as ANN outputs. molecular descriptor [25]. Since ANNs compute the output as a sum
of nonlinear transformations of linear combinations of the inputs,
Artificial Neural networks (ANNs) sensitivity shows the percentage contribution of a corresponding input
ANNs are biologically inspired computational model designed to to the output value and reveals the effect that a change in that particular
simulate the way in which the human brain processes information [23]. input has on output. Hence, inputs with low sensitivity are considered
ANNs are composed of many individual processing units or artificial to only have a minimal contribution to the model being analysed so
neurons which are extensively inter-connected with connection weights were eliminated from subsequent models.
to form a network. They collect their knowledge by detecting patterns Molecular descriptors with sensitivities less than one were
and relationships in inputs and outputs and learn from experience from sequentially removed after each automated network run until models
previously seen data, rather than using a pre-designed equation in a that contained only molecular descriptors of relative importance for
model. Artificial neurons are typically organized into an input layer, assessed wine quality parameters were developed.
one or more hidden layers, and an output (prediction) layer. They
function by linking the input neurons (i.e. spectral intensity/peak) Results and Discussion
to output neurons (i.e. measured wine quality indicator), through a
Fourier Transform Infrared (FT-IR) spectroscopy is a non-
set of connections with adjustable strengths (weights). The standard
destructive analytical technique that provides structural information
supervised network architectures (multilayer perceptrons and radial on molecular features of a large range of compounds.
basis functions) are models in which connection weights and the
number of hidden neurons are adjustable parameters that are optimized In general, no two wine samples will show exactly the same IR
during the learning phase. This is performed using the training and absorption pattern, thus a fingerprint IR spectrum can be produced,
validation sets of compounds. Training is performed iteratively, such unique to each wine. However, the common constituents of wine (i.e.
that as it progresses, the ANN will generate a more accurate output and water and ethanol) produce IR absorption bands, which may disguise
establish a (linear or non-linear) relationship between input and output the characteristic IR vibrations of phenols. This is due to the fact that
data. Following this, true predictive ability of the model can be tested water, ethanol and organic acids absorb in the same MIR region as
using an independent set of compounds, validation set. phenols. To eliminate this interference, a 12% (v/v) ethanol in water
spectrum was subtracted from the wine spectra used in this work.
Network training and design
All wine samples including those from the Bendigo region, NSW
Averaged spectral intensities, variety of grape (Merlot, Cabernet, and wines ranging over different vintages gave rise to similar spectra
Shiraz, Semillon, Sauvignon, Malbec), wine barrel (French oak, patterns (Figure 1). Several absorption bands were identified including
American oak, mixed, tank, unknown), and production year were used those within the region between 800-1750 cm-1 which are categorized
as inputs and calculated concentrations of total phenols expressed as as C=C-C aromatic ring stretches (1580-1615 cm-1; 1450-1510 cm-
Gallic Acid Equivalents (GAE g/L), acetic acid concentration (g/L), 1
) while IR bands in the area from 820 to 760 cm-1 can be attributed
total acidity (g/L), alcohol content (%), and measured pH were used as
outputs in the ANN. The most straightforward approach was used to
build the ANN model. Input/output data sets were were automatically
randomised into training (60%), testing (20%) and validation (20%)
subsets. The training set was used for learning and to fit network
parameters (weights) while the testing set was used to optimise the
network topology and avoid over fitting. An external validation
subset was used to assess the predictive performance (generalization)
of a developed neural network. The extent of training was monitored
internally by the ANN program and was stopped when the Root Mean
Square (RMS) error failed to improve during training cycles and the
testing RMS error started to increase. A number of networks with
different topologies were trained and tested to determine the optimum 4000 3500 3000 2500 2000 1500 1000 500
topology for the dopamine receptor data and back-propagation Multi Wavenumbers (cm-1)

Layer Perceptron (MLP) with one hidden layer selected due to its Figure 1: FT-MIR spectra for the different wines investigated.

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 4 of 8

Calculated are within the range of 1200 and 950 cm-1, more specifically the
Total
Wine sample
Total acidity
pH (±0.3)
acetic acid
phenols peaks observed at 1157, 1107, 1065, 1014 cm-1. The stretch vibration
(g/L) concentration of the C=O group is around 1700 cm-1. Peaks at 1618 and 1407 cm-1
(GAE* g/L)
(g/L)
corresponds to symmetrical and asymmetrical stretching vibration for
1 6.75 3.51 0.13 2.65
the carboxyl ion (COO-) indicating the existence of carboxylic acid,
2 5.24 3.74 0.27 2.31
ester, or carbonyl groups [29]. Peaks located in the region of 1450 to
3 6.73 3.38 0.37 2.53
1410 cm-1 originate from symmetric stretching vibration of C-O, and
4 4.49 3.67 0.72 1.83
those around 1500 cm-1 can be assigned to C-C stretching in rings [1].
5 6.18 3.66 1.65 1.70
6 5.99 3.53 1.12 1.21 In the same way, the IR peaks that are observed at approximately
7 6.83 3.54 0.24 0.17 1300 and 1150 cm-1 can be assigned to S=O stretch present in a sulfate
8 7.02 3.25 0.31 2.18 group Peaks at 1350 and 1175 cm-1 for sulfonates and peaks in the area
9 6.56 3.45 0.15 0.36 between 1000-750 cm-1 can be assigned to S-O [30].
10 5.68 3.55 0.52 1.52
11 5.83 3.46 0.15 1.65
Liquid sulfur dioxide shows IR bands at 530, 1142 and 1330 cm-1.
12 5.49 3.50 1.12 1.75
For the sulphuryl compounds such as sulphuric acid, the SO2 bend is
13 4.45 3.88 1.17 1.44
in the range 500-600 cm-1 [31]. The IR absorbance at 1000 cm-1 may
14 6.31 3.41 0.21 1.65
indicates the existence of PO43- [32]. The presence of polyphenolic
15 6.18 3.38 0.27 1.52
compounds is indicated by characteristic bands for gallic acid at 669,
16 5.2 3.77 1.54 2.42
763, 1025, 1100 and 1654 cm-1, tannic acid at 669, 860, 1172, 1511 and
17 7.04 3.69 1.44 2.46
1627 cm-1 [4] (Figure 1).
18 5.93 3.38 1.3 2.15 Four separate nonlinear ANN models were developed to correlate
19 6.37 3.42 0.02 1.98 the recorded IR spectra, variety of grape, wine barrel type and
20 4.63 3.48 0.52 3.06 production year, with the experimentally measured total polyphenolic
*GAE=Gallic acid equivalent content, total and volatile acidity, pH and alcohol content in the
Table 2: Total acidity, pH, acetic acid concentration, and total phenolic concentration
selected wine samples (Table 2). The strength of the correlation between
for the wine samples. experimental and predicted data was assessed by the accuracy of the
predicted values for the validation data sets (Table 3).
Correlation
The developed ANN models confirmed that the characteristic IR
ANN
Training Testing Validation Average bands, grape variety, year, wine type and wine barrel used for aging are
architecture*
Acetic acid directly related to the experimentally determined amounts of phenols,
concentration 0.960 0.998 0.924 0.898 25-1-1 total acidity, acetic acid content, pH, and alcohol content. Good linearity
(g/L) (Table 3) of the ATR-FTIR method indicates that this technique can be
Alcohol content 0.914 0.990 0.789 0.942 147-1-1 used to assess the quality of wine accurately and specifically.
Total polyphe-
0.999 0.904 0.925 0.918 186-5-1
nols (g/L GAE) Total polyphenols
Total acidity
0.954 0.999 0.801 0.898 143-1-1 Not surprisingly the ANN model for the polyphenolic content
(g/L)
*Number of inputs-hidden neurones-number of outputs
(186-5-1) has the wine barrel as the most important input contribution
in addition to the peaks at 1870,3740 2200 and 530-550 due to alcohol,
Table 3: Correlation data for the four developed ANN models. carbonyl, alkyne, nitriles and vibrations in the fingerprint region. The
to ring vibrations [26]. Furthermore, peaks between 670-900 cm-1 composition of untreated wine samples consists of a complex mixture
can be attributed to aromatic C-H out of plane (750-1000 cm-1) and of various phenolic compounds and hence classification of each
in plane bending (950-1225 cm-1) [27]. The IR regions of significant specific phenol by their spectral peak profile can be difficult. Never
importance to this study were from 1542 to 965 cm-1, usually referred the less, the distinctive absorption bands within the finger print region
to as the “fingerprint” region and various IR bands, including those of the IR spectrum can be assigned to particular functional groups
corresponding to the vibration of the C-O, C-C, C-H and C-N bonds, present in several phenolic compounds. Moreover, wine consists
occurs in this region [1,28]. This area provides important information of several components including water (80 to 90 percent), alcohol,
regarding organic compounds such as sugars, alcohols and organic sugar, carboxylic acids, tannins, polyphenols, amino acids, vitamin
acids present in the sample. The distinct absorbance peaks in the wave C, inorganic components and numerous fragrance ingredients [3].
number regions 3626-2970 cm-1 and 1716-1543 cm-1, are the result Therefore, acknowledging their characteristic IR functional group
of the absorbance of water [29]. Other absorption bands of interest absorption peak(s) can assist in more accurate determination of the
involved those at 1044 and 1085 cm-1, which are indicative of an alcohol phenolic compounds present.
functional group. The 1382 cm-1 absorption band attributes to the O-H Red wines are rich in polyphenolic substances, mainly flavanols
in plane deformation in polyphenols [3]. The cyclic nature of the ether and anthocyanins, which contribute to the sensory properties of wines,
was reflected by the peaks located at 1283-1247cm-1 range and as well at
such as colour, taste, astringency and flavour [15,33]. Furthermore, the
1158 cm-1, which was produced by the aromatic C-O bond stretching.
total phenol content directly correlates with the antioxidant activity
The 1739 cm-1 absorption band may attributed to the carbonyl group,
of wines [34]. The antioxidant activity of wine polyphenols has been
C=O of the galloyl unit on epicatechin gallate. The deformation
demonstrated in many studies and they act via various chemical
vibration of the carbon–carbon bonds in the phenolic groups adsorb in
pathways (i.e. as free radical terminators, singlet oxygen quenchers,
the region of 1500-1400 cm-1 [2].
and chelators of metal ions) [35,36]. Inhibition of low-density
IR absorption due to the presence of sugar functional groups lipoprotein oxidation, inhibition of platelet aggregation, and anti-

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 5 of 8

Furthermore, the total phenolic content increases as the alcohol

content of the wine increases (Figure 3). Optimal antioxidant activity is
0.10
achieved at an alcohol concentration of approximately 15% v/v. Above
this range, however, antioxidant activity declines. Results show that the
Peak intensity at 1247 cm-1

0.08
high alcohol content wines in the group (wine samples 4, 5, 16, 17, and
0.06 20), also have the highest overall antioxidant concentrations. However,
wine sample 8, with a much lower alcohol content of 13.2% v/v had the
0.04 highest antioxidant concentration of all of the wines in the group. This
result however, may not be reliable as it was an outlier from the general
0.02 trend and may have been subject to experimental error. Further testing
of a wider range of wine samples with varying alcohol content is needed
0.00
10 11 12 13 14 15 16 in order to confirm the reliability of this correlation. Moreover, the
Alcohol content (% v/v) true antioxidant activity of a wine sample in vivo cannot be accurately
determined unless physiological testing is also performed (Figure 3).
Figure 2: IR absorption at the characteristic peak for flavonoids at 1247
cm-1 (C-O stretching frequency of esters), versus alcohol content. The phenolic composition of red wine depends on many factors,
such as grape variety, wine maturation and ageing. During wine
maturation and aging the polyphenolic components extracted from
3.5
the grapes undergo various reactions, the most important being
condensation reactions between anthocyanins and flavonoids, and
3.0
reactions involving other wine components, such as acetaldehyde,
2.5 pyruvate, hydroxycinnamic acids and vinyl phenols, which give rise
Total phenols (g/L GAE)

to more stable oligomeric or polymeric pigments [42,43]. The rate

2.0 and extent of these reactions is influenced by various parameters,
including the initial concentration of reactant species, pH of the
1.5 medium, temperature, ageing conditions, oxygen availability and the
concentration of antioxidant agents such as sulphites [44-46]. It is also
1.0
known that the effect of the type of wine barrel used is a contributing
factor to the total phenolic content of red wine (Figure 4).
0.5
The aging of wines in wine barrels can potentially provide a modest
0.0
12.0 12.5 13.0 13.5 14.0 14.5 15.0 15.5 16.0 contribution to the phenolic content of wine. The oak barrels contain
Alcohol (% v/v)
gallotanins and ellagitannins which leach into the wine from the barrels
resulting in an increase in the total tannin levels found in the wine [47].
Figure 3: Plot of total phenolic content (GAE g/L) versus alcohol
concentration (% v/v). For this reason, we need to consider the type of oak barrel used in wine
storage as an important factor in determining the overall level of total
phenolic compounds found in wine. Results indicate that variations
inflammatory properties by red wine phenolics [9,37] may be a major in wine barrel oak type (i.e. French or American oak) will affect the
factor contributing to the health benefits of red wine and the decreased amount of tannins extracted into the wine. (Figure 4) indicates that
risk of cardiovascular diseases despite a high-fat diet in certain French French oak based wines achieved a small yet higher content of total
populations [38]. The main component in red wine that is believed polyphenols than American oak based wines. According to studies,
to provide cardiovascular protection is resveratrol (3,4,5-trihydroxy-
French oak contains an overall higher concentration of tannins than
trans-stilbene) and together with polyphenols are produced exclusively
American oak [48]. For this reason several wineries opt for more
in the vine leaf epidermis and grape skin [39]. The resveratrol content of
wine is related to the length of time the grape skins are present during
the fermentation process. It was found that the synergistic contribution
due to the presence of both phenolic and polyphenolic compounds, 4.5
might be more important for antioxidant activity than any specific
4.0
Total polyphenolic content (g/L GAE)

phenolic concentration [40]. Polyphenolic compounds are found in the

skin and seeds of grapes. Red wines have a considerably higher amount 3.5
of polyphenolic compounds than white wines, because the skins are 3.0
removed earlier during white-wine production, lessening the amount
2.5
that is extracted [6]. When wine is made, the alcohol produced by the
fermentation process dissolves the polyphenols contained in the skin 2.0

and seeds. 1.5

Based on Raman measurements of pure flavones and related 1.0

compounds, the intense band at 1247 cm-1 can be attributed to 0.5
flavonoid-type compounds [41]. The intense band at 1247 cm-1, that
could be assigned to bending of the OH groups coupled to C-H in plane French American Mixed Unknown Tank
bend in flavonoids, correlates well with the alcohol content found in Barrel type
a wine sample (Figure 2), but does not relate with total polyphenolic Figure 4: Total polyphenolic content as a function of wine barrel.
content, acidity and pH [16].

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 6 of 8

Titratable or total acidity

The most important descriptors that were included in the model for
the total acidity were bends attributed the broad stretch of carboxylic
acid O-H (2500-3000 cm-1), carbonyl group stretching vibrations
(1670-1820 cm-1), mono substituted benzene ring bend (strong band
at 690 cm-1), nitrile (2210-2260 cm-1), wine type and wine barrel (oak
Total acidity (g/L)

type). The total acidity of the wine is a measure of all types of acids
present, i.e. inorganic acids, such as phosphoric acid, and organic acids
where tartaric acid and malic acid are the predominate components,
contributing to around 90% of the overall total acidity of the wine.
There are also amino acids whose contribution to titratable acidity is
not very well known. Often the total acidity is quantified simply as the
measure of the total amount of tartaric acid present [49].
Organic acids (i.e. tartaric acid) make major contributions to the
taste, feel and colour of the wine. More importantly their preservative
Ac properties also enhance wines microbiological and physicochemical
eti
c aci stability. Red wines are stable at lower acidity, due to the presence of
d( phenols which enhance acidity, balance the sweet taste of the alcohols
g /L pH and help to maintain stability throughout the aging process. Acidity also
)
greatly influences the taste and colour of wine, and is important when
Figure 5: The relationship between acetic acid concentration assessing the general quality to the wine [50]. Grapes grown in warmer
(volatile acidity), pH and total acidity. climates tend to have higher sugar content and lower acidity, compared
with colder climate grapes. In general the total acidity of wine varies
expensive French oak in order to help release phenols and tannins into from 0.4% v/w to 1.0% w/v (4-10 g/L), but usually in red wines the
the wine in order to improve flavor and structure. Wines aged in French acidity is from 0.6% w/v to 0.8% w/v (6–8 g/L). The measured total
and American oak both showed higher absorbance readings than wines acidity of the wine samples in this work is within an acceptable range of
that have been stored in fermenting tanks. 4.45 to 7.04 g/L. A total of 8 different wines had acetic acid levels greater
than 0.5 g/L, with 2 of them above 1.5 g/L. A high level of spoilage is due
As expected, our results show that the total phenolic content was to the concentration of acetic acid being anything above 1.5 g/L, but for
higher in the red wine samples. Furthermore, despite a spread in the good quality wine, the level should be less than 0.5 g/L [51].
amounts of the phenolic content present in the different wine varieties
and vineyards, cabernet sauvignon wines had higher levels of phenolics Wine pH
compared to merlot wines.
As a wine quality parameter, pH is of equal or greater significance
Acetic acid content (Volatile acidity) than the titratable total acidity. It affects wine colour, taste, oxidation-
reduction potential, ratio of free to bound sulphur dioxide, and the
The ANN model correlating IR spectral bends with acetic acid
extent of iron phosphate cloudiness present in the wine [52]. The pH
content was the simplest with only 25 input descriptors, mostly related
and total acidity of a wine follows an inverse proportional relationship,
to carbonyl stretching vibration between 1650-1750 cm-1 and peaks in
where an increase in pH correlates to a lower total acidity. However,
the fingerprint region (500-550 cm-1) in addition to wine variety (red
while there is a general relationship between total acidity and pH, these
wine versus white wine) and type of the wine barrel used for aging.
parameters are not directly related to each other. The association is
Volatile acids consist of a number of low molecular weight fatty acids
regarded as an empirical relationship between the pH and the ratio of
[49] with acetic acid as the predominant acid. High levels of acetic acid
are indicative of spoilage, mostly from Acetobacter. A small amount potassium bitartrate to total tartaric acid. Note that pH only measures
of acetic acid is normal, but should not exceed 0.3 g/L. Higher levels the free hydrogen ions in solution, while titratable acidity measures the
of acetic acid; especially 0.6-0.9 g/L is enough to provide a sensory concentration of all of the available hydrogen ions, both those free in
indication of spoilage. At the level of 1.5 g/L the wine has essentially solution and those bound to undissociated acid molecules. Most of the
turned into vinegar. It can be seen from the 3D surface plot (Figure wines samples in this study came from the central Victorian region,
1), that there is a correlation between pH, total acidity and acetic acid which is classified as a warm climate wine growing area. These wines
content, presenting optimal aspects for wine quality. As discussed are expected to have lower levels of acids and higher pH values, and
earlier, in order to have a better quality wine, ideally the wine would have are known for deep colour, low tannins and high levels of acid (which
a higher content of total acidity and a lower acetic acid concentration. is unusual for a warm climate red grape). As the pH shifts to the lower
This can be seen at a pH of approximately 3.6 where spoilage is at a end of the scale, the colour of the wine shifts to a more intense red form.
low (acetic acid < 0.75 g/L) yet total acidity is not affected. Therefore With the taste of wine, pH values around 4.0 results in a flat taste while
according to (Figure 1), a pH of 3.6 and a total acidity of 7.0 g/L would a pH of less than 3.0 results in a tart and sour taste. On average, the pH
optimise wine quality in relation to these parameters. It is important of red wine should not exceed 3.6 [53], with the optimum pH range
to note the wines with a pH and total acidity in this region also have for wine between 3.2-3.6. Although according to literature, there is no
higher a phenolic content (and antioxidant capacity) than the other direct correlation proven between pH, volatile acidity and total acidity,
wines investigated. At higher pH levels, acetic acid content becomes there is a more general non-linear relationship between the total acidity,
more significant thereby affecting wine quality. pH and volatile acidity, which can be seen in Figure 5 [54].

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 7 of 8

Conclusion 14. Kennedy JA, Matthews MA, Waterhouse AL (2000) Changes in grape seed
polyphenols during fruit ripening. Phytochemistry 55: 77-85.
Parameters used to assess grape and wine quality such as total and 15. Robichaud JL, Noble AC (1990) Astringency and bitterness of selected
volatile acidity, pH, alcohol, and antioxidant content were determined phenolics in wine. Journal of the Science of Food and Agriculture 53: 343-353.
for each wine sample using conventional analytical techniques. Each 16. Canamares MV, Lombardi JR, Leona M (2009) Raman and surface enhanced
of these parameters affects the overall quality of the wine, either raman spectra of 7-hydroxyflavone and 3’,4’-dihydroxyflavone. e-Preservation
contributing to taste, colour, protective and preservative properties, and Science 6: 81-88.
potential health benefits. 17. Bauer R, Nieuwoudt H, Bauer FF, Kossman J, Koch KR, et al. (2008) FTIR
spectroscopy for grape and wine analysis. Anal Chem 80: 1371-1379.
The total acidity of the wines ranged from 4.45 to 7.04 g/L, pH
18. Urtubia A, Perez-Correa JR, Pizarro P, Agosin E (2008) Exploring the
from 3.25 to 3.88, acetic acid concentration from 0.02 to 1.54 g/L, and applicability of MIR spectroscopy to detect early indications of wine fermentation
polyphenolic content from 0.17 to 3.06 g/L. Within the FTIR fingerprint problems. Food Control 19: 382-388.
region obtained for each wine sample, several absorption bands were
19. Jensen JS, Egebo M, Meyer AS (2008) Identification of spectral regions
identified as functional groups representative of those present in for the quantification of red wine tannins with fourier transform mid-infrared
phenolic compounds. A plot of the values predicted by the validated spectroscopy. J Agric Food Chem 56: 3493-3499.
ANN models showed excellent correlation with the experimentally 20. Versari A, Boulton RB, Parpinello GP (2006) Effect of spectral pre-processing
measured values (r=0.898-0.942) for acetic acid concentration, alcohol methods on the evaluation of the color components of red wines using Fourier-
content, total phenols, and total acidity. Experimental data collected transform infrared spectroscopy. Italian Journal of Food Science 4: 423-432.
from the wine samples showed that the phenolic content of the wines 21. Versari A, Parpinello GP, Scazzina F, Del Rio D (2010) Prediction of total
may be influenced by the type of oak used for aging the wines. Also antioxidant capacity of red wine by Fourier transform infrared spectroscopy.
Food Control 21: 786-789.
there appeared to be a direct correlation between the alcohol content
and the total phenolic content of the wines. FTIR spectroscopy proved 22. Fuller WA (1996) Introduction to statistical time series. John Wiley & Sons, Inc.,
New York, USA.
to be a promising technique that provides a rapid method in the quality
assessment of wine, for determining acetic acid concentration, alcohol 23. Agatonovic-Kustrin S, Beresford R (2000) Basic concepts of artificial neural
network (ANN) modeling and its application in pharmaceutical research. J
content, total polyphenols, and total acidity. Moreover, this method
Pharm Biomed Anal 22: 717-727.
uses small sample volumes (a few milliliters) with almost no sample
preparation required. 24. Agatonovic-Kustrin S, Turner JV, Glass BD (2008) Molecular structural
characteristics as determinants of estrogen receptor selectivity. J Pharm
References Biomed Anal 48: 369-375.

1. Movasaghi Z, Rehman S, Rehman IU (2007) Raman spectroscopy of biological 25. Turner JV, Maddalena DJ, Cutler DJ, Agatonovic-Kustrin S (2003) Multiple
tissues. Applied Spectroscopy Reviews 42: 493-541. pharmacokinetic parameter prediction for a series of cephalosporins. J Pharm
Sci 92: 552-559.
2. Nakano Y, Takeshita K, Tsutsumi T (2001) Adsorption mechanism of hexavalent
chromium by redox within condensed-tannin gel. Water Res 35: 496-500. 26. Lin-Vien D, Colthup NB, Fateley WG, Grasselli JG (1991) The handbook of
Infrared and Raman characteristic frequencies of organic molecules. Academic
3. Ozacar M, Soykan C, Sengil IA (2006) Studies on synthesis, characterization, Press, New York, USA.
and metal adsorption of mimosa and valonia tannin resins. Journal of Applied
Polymer Science 102: 786-797. 27. Yadav BS, Tyagi SK, Seema (2006) Study of vibrational spectra of 4-methyl-
3-nitrobenzaldehyde. Indian Journal of Pure & Applied Physics 44: 644-648.
4. Kannan RRR, Rajasekaran A, Perumal A (2011) Fourier transform infrared
spectroscopy analysis of sea grass polyphenols. Current Bioactive Compounds 28. Smith BC (1999) Infrared spectral interpretation: A systematic approach. 1st
7: 118-125. edition, CRC Press LLC, Florida, USA.

5. Ferrieres J (2004) The French paradox: lessons for other countries. Heart 90: 29. Zhang Y, Chen J, Lei Y, Zhou Q, Sun S, et al. (2010) Discrimination of different
107-111. red wine by Fourier-transform infrared and two-dimensional infrared correlation
spectroscopy. Journal of Molecular Structure 974: 144-150.
6. Kopp P (1998) Resveratrol, a phytoestrogen found in red wine. A possible
explanation for the conundrum of the ‘French paradox’? Eur J Endocrinol 138: 30. Flor S, Tripodi V, Contin M, Dobrecky C, Lucangioli S (2012) Spectroscopic
619-620. approach of the association of heparin and its contaminant and related
polysaccharides with polymers used in electrokinetic chromatography. Journal
7. Goldberg DM, Hahn SE, Parkes JG (1995) Beyond alcohol: beverage of Chemical and Pharmaceutical Research 4: 972-979.
consumption and cardiovascular mortality. Clin Chim Acta 237: 155-187.
31. Anderson A, Savoie R (1965) Raman spectrum of crystalline and liquid SO2.
8. Just JR, Daeschel MA (2003) Antimicrobial effects of wine on Escherichia coli Canadian Journal of Chemistry 43: 2271-2278.
O157:H7 and Salmonella typhimurium in a model stomach system. Journal of
Food Science 68: 285-290. 32. Yu D, Wu C, Kong Y, Xue N, Guo X, et al. (2007) Structural and catalytic
investigation of mesoporous iron phosphate. J Phys Chem C 111: 14394-
9. Frankel EN, Waterhouse AL, Teissedre LP (1995) Principal phenolic 14399.
phytochemicals in selected California wines and their antioxidant activity in
inhibiting oxidation of human low-density lipoproteins. J Agric Food Chem 43: 33. Eiro MJ, Heinonen M (2002) Anthocyanin color behavior and stability during
890-894. storage: effect of intermolecular copigmentation. J Agric Food Chem 50: 7461-
7466.
10. Soleas GJ, Diamandis EP, Goldberg DM (1997) Wine as a biological fluid:
history, production, and role in disease prevention. J Clin Lab Anal 11: 287-313. 34. De Beer D, Harbertson JF, Kilmartin PA, Roginsky V, Barsukova T, et al. (2004)
Phenolics: A comparison of diverse analytical methods. Am J Enol Vitic 55:
11. Ikigai H, Nakae T, Hara Y, Shimamura T (1993) Bactericidal catechins damage 389-400.
the lipid bilayer. Biochim Biophys Acta 1147: 132-136.
35. Kinsella JE, Frankel EN, German B, Kanner J (1993) Possible mechanisms
12. Jayaprakasha GK, Selvi T, Sakariah KK (2003) Antibacterial and antioxidant for the protective role of antioxidants in wine and plant foods. Food technology
activities of grape (Vitis vinifera) seed extracts. Food Research International 47: 85-89.
36: 117-122.
36. Packer L, Hiramatsu M, Yoshikawa T (1999) Antioxidant food supplements in
13. Vaquero MJR, Alberto MR, de Nadra MCM (2007) Influence of phenolic human health. Academic Press, London, UK.
compounds from wines on the growth of Listeria monocytogenes. Food Control
18: 587-593. 37. Saito M, Hosoyama H, Artiga T, Kataoka S, Yamaji N (1998) Antiulcer activity
of grape seed extract and procyanidins. J Agric Food Chem 46: 1460-1464.

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal
Citation: Agatonovic-Kustrin S, Morton DW, Yusof AP Md (2013) The Use of Fourier Transform Infrared (FTIR) Spectroscopy and Artificial Neural Networks (ANNs) to
Assess Wine Quality. Mod Chem appl 1: 110. doi:10.4172/2329-6798.1000110

Page 8 of 8

38. Renaud S, de Lorgeril M (1992) Wine, alcohol, platelets, and the French 46. Romero C, Bakker J (2000) Effect of storage temperature and pyruvate on
paradox for coronary heart disease. Lancet 339: 1523-1526. kinetics of anthocyanin degradation, vitisin A derivative formation, and color
characteristics of model solutions. J Agric Food Chem 48: 2135-2141.
39. Liu BL, Zhang X, Zhang W, Zhen HN (2007) New enlightenment of French
Paradox: resveratrol’s potential for cancer chemoprevention and anti-cancer 47. Waterhouse AL (2002) Wine phenolics. Ann N Y Acad Sci 957: 21-36.
therapy. Cancer Biol Ther 6: 1833-1836.
48. Fernandez K, Agosin E (2007) Quantitative analysis of red wine tannins using
40. Lin YT, Vattem D, Labbe RG, Shetty K (2005) Enhancement of antioxidant Fourier-transform mid-infrared spectrometry. J Agric Food Chem 55: 7294-
activity and inhibition of Helicobacter pylori by phenolic phytochemical-enriched 7300.
alcoholic beverages. Process Biochemistry 40: 2059-2065.
49. Ough CS, Amerine MA (1988) Methods of analysis of musts and wines. 2nd
41. Yamauchi S, Shibutani S, Doi S (2003) Characteristic raman bands for edition, John Wiley & Sons, New York, USA.
artocarpus heterophyllus heartwood. Journal of Wood Science 49: 466-468.
50. Mataix E, de Castro MDL (1999) Sequential determination of total and volatile
42. Bakker J, Timberlake CF (1997) Isolation, identification, and characterization acidity in wines based on a low injection-pervaporation approach. Analytica
of new color-stable anthocyanins occurring in some red wines. J Agric Food Chimica Acta 381: 23-28.
Chem 45: 35-43.
51. Zoecklein BW, Fugelsang KC, Gump BH, Nury FS (1999) Wine analysis and
43. Monagas M, Nunez V, Bartolome B, Gomez-Cordoves C (2003) Anthocyanin- production. Chapman & Hall, New York, USA.
derived pigments in graciano, tempranillo, and cabernet sauvignon wines
produced in Spain. Am J Enol Vitic 54: 163-169. 52. Peynaud E, Blouin J (1996) The taste of wine: The art science of wine
appreciation. John Wiley and Sons, New York, USA.
44. Schwarz M, Wabnitz TC, Winterhalter P (2003) Pathway leading to the formation
of anthocyanin-vinylphenol adducts and related pigments in red wines. J Agric 53. Jackson RS (2000) Wine science: Principles and Applications. 3rd edition,
Food Chem 51: 3682-3687. Academic Press, San Diego, USA.

45. Romero C, Bakker J (2000) Anthocyanin and colour evolution during maturation 54. Boulton R (1980) The relationships between total acidity, titratable acidity and
of four port wines: Effect of pyruvic acid addition. Journal of the Science of Food ph in wine. American Journal of Enology and Viticulture 31: 76-80.
and Agriculture 81: 252-260.

Mod Chem appl Volume 1 • Issue 4 • 1000110

ISSN: 2329-6798 MCA, an open access journal