Classification of commercial apple juices based on multivariate analysis of their chemical profiles

ABSTRACT

Currently, a wide range of differently processed apple juices is available on the market. In this study the quality of commercial apple juices from four product categories was evaluated on the basis of their chemical profiles (total soluble solids, pH, titratable acidity, ratio of soluble solids to acidity, sugars, total phenolics content, and antioxidant activity) using multivariate methods. Principal component analysis has revealed, that the chemical parameters, such as titratable acidity, ratio of soluble solids to acidity, and pH (PC1) and sugars and total soluble solids (PC2) appears to be the parameters most differentiating the samples. Five classes of juices with similar chemical composition were detected using hierarchical cluster analysis. The exploratory analysis of the overall chemical profiles revealed that the juices clear from concentrate, cloudy not from concentrate and freshly squeezed, were easily distinguishable due to their unique properties. In contrast, cloudy juices from concentrate showed properties similar to juices of other classes. The classification based on k-nearest neighbors method had high sensitivity and low classification error for clear juices from concentrate and cloudy not from concentrate. The classification failed for the cloudy juices from concentrate.

KEYWORDS:

• Chemometrics
• Authentication
• Principal component analysis
• Hierarchical cluster analysis
• k–nearest neighbors

Introduction

The apple juice is widely consumed and appreciated due to its specific pleasant taste and high nutritional value. An apple is a fruit with high content of phenolic compounds, vitamins, minerals, dietary fibre and high antioxidant capacity.^[1–3] The chemical composition of apple juices is strongly influenced by a variety of factors, including properties of the raw material (such as maturity, cultivar, growing region, cultivation practices, climate, water stress, rootstock, pest resistance, storage conditions, and others) and the processing technology.^[4]

Currently a wide variety of differently processed apple juices is available on the market. Clear juices reconstituted from concentrate (FC) are the most common. However, the native pectin, insoluble solids and phenolic compounds are enzymatically degraded during the production of the clarified juices.^[5–7] This process negatively affects the presumable health-promoting functionality of the apple juices associated with the high content of pectin and phenolic compounds. Some research suggests that consumption of cloudy juices might provide more favorable effects on human health, as compared to clear apple juices.^[8,9] Therefore, technologies for the production of cloudy juices are being developed and improved.^[10] The cloudy juices available on the market include those reconstituted FC enriched with pulp and naturally cloudy pasteurized juices not-from-concentrate (NFC). Another segment includes freshly squeezed unpasteurized juices. These preserve the sensory properties of raw apples, fulfilling the consumer demand for minimally processed food; however, their shelf life when stored chilled is a only a few days.

The overall quality of the fruit-based products is usually determined by a variety of chemical, physical, and sensory parameters. Multivariate methods are used for the analysis of such data.^[11,12] Chemometric techniques have been successfully used to monitor or assure the quality of many food products including fruits.^[13–16] Application of chemometrics to the studies of the fruit juices has been recently reviewed.^[17] It was demonstrated that the multivariate methods are useful to study diverse aspects of the juice quality. For example, chemometric techniques were used to classify juices according to different criteria, to detect adulteration, to authenticate, to assess origin, and to explore sensory and chemical properties.^[17,18]

In studies of the apple juices multivariate methods were used for classification of juices according to the apple cultivar and geographical origin on the basis of their polyphenol composition,^[19] for correlation of the analytical and sensory data,^[20–22] for comparison of the chemical composition of differently processed juices from various apples cultivars.^[23] Most of the articles published on the compositional data mainly investigated the laboratory-prepared apple juices.^{[4,19,22–26]} A much lower number studied commercial apple juices, focusing on selected aspects of quality, e.g., polyphenol content and antioxidant capacity,^[27,28] aroma profiles,^[21] and sensory profiles.^[29]

The quality of juices available on the market should be of paramount interest not only for the consumers but also for all of the bodies involved in the entire production chain. The aim of this study was to evaluate and compare the quality of the apple juices from different product categories based on their chemical properties using multivariate methods. The chemical profiles of the juices studied encompassed several parameters: total soluble solids (TSS), sucrose, D-glucose, and D-fructose content, pH, acidity, ratio of soluble solids to acidity, total phenolics, and antioxidant capacity. These parameters affect the nutritional value, sensory properties and the health-promoting function of the apple juices.^[20,24,30] Multivariate methods were used for the analysis of the overall chemical profiles. Principal component analysis (PCA) and hierarchical cluster analysis (HCA) were used for the exploratory study of the chemical data. Classification model was developed using the k-nearest neighbors (kNN) method.

Materials and methods

Apple juice samples

Ninety apple juice samples that are available on the Polish market were assessed in this study. Juices from 18 different producers were studied; all samples were taken from five different production batches. Juices were acquired between 2012 and 2014. The studied samples included juices that were produced from concentrate (FC), clear (group A: 35 samples) and pulp-enriched cloudy varieties (group B: 10 samples), and not-from-concentrate (NFC) juices, including pasteurized naturally cloudy juices (group C: 35 samples), and freshly squeezed juices (group D: 10 samples). According to the label, two cloudy juices FC, two NFC juices, and one fresh juice were enriched in vitamin C. No further systematic information about the raw material characteristics and details of respective processing procedures was available.

Physicochemical measurements

The TSS content

TSS was measured using a DR-A1’s Conbest Abbe Refractometer at 22°C, and expressed as °Brix.

Sugars

Juices were diluted in water 1:99 and analyzed using the Boehringer Mannheim enzymatic test kit (Cat. No. 10 716 260 035, R-Biopharm AG, Darmstadt, Germany). The test is based on absorbance measurements at 340 nm of NADPH, formed under specific enzymatic reactions. A Milton Roy Spectronic Genesys 2 spectrophotometer was used to measure the absorbance. The results were expressed in gram per liter of juice.

pH

The pH measurements were performed at 22°C using Seven pH meter with pH combination electrode from Mettler Toledo Compact.

Acidity

Titratable acidity (TA) was measured by titrating with 0.1 M NaOH to a pH end-point of 8.1. The results were expressed as gram of citric acid per liter of juice (g CA/L).

Total phenolic content

The total content of phenolic compounds was determined according to Singleton and Rossi^[31] with modification for measurement on 48-well microplates. This method is based on the absorbance measurement of the reaction product at 765 nm, which is used, to quantify the reduction of the Folin–Ciocalteu reagent (Sigma-Aldrich, Steinheim, Germany). In brief, 0.01 mL of each juice sample was diluted 1:2 with water and mixed with 0.05 mL of Folin–Ciocalteu reagent. After 3 min, 0.15 mL of 20 % sodium carbonate and 0.79 mL of deionized water were added and the solution was mixed well. After 120 min in the dark at room temperature, the absorbance was measured using an EpochTH microplate spectrophotometer (BioTek). The total content of phenolic compounds was expressed in mg of gallic acid equivalents per liter of juice (mg GAE/L).

Total antioxidant capacity

Antioxidant capacity was determined using the Trolox equivalent antioxidant capacity (TEAC) assay according to Re et al.^[32] as described in detail in Gliszczynska-Swigło et al.^[33] This method quantifies the loss of absorption of the ABTS˙⁺ radical cation, measured at 734 nm, with the increase of juice/Trolox concentration. The TEAC value was calculated from the linear regression coefficient of the calibration curve for six dilutions of the juice, as calibrated against the linear regression coefficient of the Trolox curve. The ABTS˙⁺ radical cation was generated by tracking the interaction of 7.7 mg of ABTS that was dissolved in 1.8 mL of deionized water and 0.2 mL of 0.0069 g/mL potassium persulphate. The cation was incubated in the dark, at room temperature for 16 h. The ABTS˙⁺ radical cation working solution was diluted with methanol to an absorbance of 0.80 at 734 nm. The absorbance was recorded 6 min after mixing 0.008 mL of juice with 0.792 mL of the ABTS˙⁺ working solution. The TEAC value was expressed in mmol of Trolox per liter of juice (mM), with Trolox being used as a standard in this assay. The measurements were performed using Milton Roy Spectronic Genesys 2 spectrophotometer. All physicochemical determinations were performed at least in triplicate.

Data analysis

Univariate statistical analysis

A one-way analysis of variance (ANOVA) was performed on the chemical data. When differences were significant, Tukey’s multiple comparison test was applied to evaluate the differences between the samples. A 5% significance level was considered. The Pearson’s correlation coefficients were calculated to analyse the linear correlation between some properties of juices.

Multivariate exploratory analysis

PCA and HCA using Ward’s method were performed on the autoscaled matrix containing 90 objects (samples) and nine variables (physicochemical parameters).

Classification method

The classification procedure consisted of the following steps: (1) selection of a training and a test set; (2) building of a model using the samples that constitute the training set and their known categories; and (3) validation of the model using an independent test set of samples.^[11]

All of the studied samples were divided into the training set (60 samples) and the test set (30 samples) using the Kennard-Stone algorithm.^[34] The training set was used for the developing classification model based on kNN method. The classification analysis was performed on the autoscaled data. The kNN is based on the determination of the distances between an unknown object and each of the objects of the training set. In kNN, k objects nearest to the unknown sample are selected and a majority rule is applied; the unknown is classified into the group to which the majority of the k objects belongs. The choice of k was optimized by calculating the prediction ability with different k values. Optimal value of k = 3 was selected by the cross-validation procedure (k giving the lowest classification error).^[11]

The classification model was evaluated on the basis of sensitivity, defined as the proportion of positive cases that were correctly classified as positive, specificity, defined as the proportion of negative cases that were classified correctly, and the misclassification error, defined as the proportion of samples that were classified incorrectly.^[11] The test set was used to evaluate the model predictive abilities for new samples. The data analyses were performed using the Statistica v. 12 (StatSoft Inc.) and Solo v. 7.5 (Eigenvector Research Inc.) software packages.

Results and discussion

Chemical profiles of the apple juices

Chemical profiles of ninety apple juice samples were evaluated. Each of the studied juices belongs to one of the four product categories: clear FC, cloudy FC, cloudy NFC, or fresh. The chemical data for individual juices are presented in Table 1. The studied juices varied significantly in chemical profiles. The dissimilarities within each of the group of juices and between juices from different product batches were observed. The characteristics of most of the studied juices were compliant with the requirements of the Code of Practice (CoP) developed by the European Fruit Juice Association,^[35] which provides reference for the control of juice quality on the European Union (EU) market. However, some individual juices did not meet the desired values. The solids content in the studied apple juices was in the range of 10.4° in fresh juice to 14.9° Brix in NFC juice. The TSS content below CoP limit were found in some clear juices FC. The composition of the individual sugars was diverse in the juices studied. The sucrose contents were from 0.8 to 39.6 g/L, the lowest content was found in NFC juice and the highest in fresh juice. Sucrose content below the limit of CoP was detected in several samples, all from cloudy NFC juice category, a few juices also exceeded the limit. Glucose content was in the range of 11.6 g/L in fresh juice to 38.1 g/L in NFC juice, and contents below the recommended limit were found in some samples of fresh juice, and some of juices from others categories slightly exceeded the CoP maximum. In accordance with literature data, fructose was the main sugar in all studied apple juices.^[4,23,24] Fructose content was on average 68.9 g/L, and ranged from 47.9 g/L in clear juice FC to 104.1 g/L in cloudy FC juice. The content of fructose exceeded the upper limit of 85.0 g/L in some juices from different categories. The results for individual sugars contents are in agreement with data previously reported.^[4,23,24] The sugars content in apple juices depends on the cultivar and to some extent on the technology of processing.

Table 1. Chemical characteristics of the individual apple juices from five different batches (a–e): total soluble solids (TSS), sucrose, glucose, fructose, pH, titratable acidity (TA), ratio of TSS/TA, total phenolics, total antioxidant capacity, TEAC, and requirements of the AIJN Code of Practice (CoP) developed by the European Fruit Juice Association.

	TSS(°Brix)	Sucrose(g/L)	Glucose(g/L)	Fructose(g/L)	pH	TA(g CA/L)	TSS/TA	Total phenolics(mg GAE/L)	TEAC(mM)
Clear FC
1a	11.5	15.6	26.2	82.2	3.4	5.6	2.1	299	2.3
1b	11.1	17.2	27.9	79.6	3.3	5.8	1.9	398	2.2
1c	10.6	12.9	21.5	58.9	3.4	6.5	1.6	329	2.5
1d	11.4	16.1	18.8	57.1	3.2	5.6	2.0	298	2.9
1e	11.2	11.2	25.1	65.7	3.4	5.8	1.9	295	2.4
2a	11.9	10.5	31.3	78.8	3.4	5.1	2.3	232	1.3
2b	11.3	13.7	23.3	83.3	3.3	5.0	2.3	205	1.4
2c	12.0	17.8	31.3	82.7	3.4	6.0	2.0	193	1.5
2d	11.6	11.1	22.4	57.4	3.2	5.0	2.3	351	2.4
2e	11.1	12.7	22.5	78.9	3.5	4.9	2.3	256	1.9
3a	11.7	6.1	21.0	57.3	3.3	5.2	2.3	171	1.3
3b	11.2	21.6	26.7	74.6	3.3	5.6	2.0	418	2.5
3c	11.3	8.0	20.4	52.9	3.1	5.3	2.1	369	2.1
3d	11.3	18.5	17.3	55.4	3.2	5.0	2.2	425	3.9
3e	11.3	23.3	22.5	63.3	3.4	5.2	2.2	116	1.1
4a	11.8	8.3	27.0	61.2	3.5	4.6	2.6	193	2.3
4b	10.9	13.7	30.0	82.3	3.4	4.7	2.3	297	2.4
4c	11.1	8.8	32.7	87.2	3.4	5.0	2.2	449	2.7
4d	11.3	15.3	20.7	62.7	3.2	4.9	2.3	359	2.7
4e	11.0	10.1	25.4	64.4	3.4	5.2	2.1	280	2.2
5a	11.3	11.1	33.7	68.6	3.4	5.2	2.2	292	2.3
5b	11.1	37.4	25.7	61.9	3.4	5.6	2.0	458	2.8
5c	11.4	7.3	23.8	53.4	3.4	5.8	2.0	602	3.6
5d	11.7	11.4	21.4	49.3	3.2	5.6	2.1	447	4.6
5e	11.1	10.2	16.4	58.0	3.4	5.2	2.1	248	2.0
6a	11.9	21.9	42.4	86.2	3.4	5.8	2.1	467	3.1
6b	10.7	20.8	23.1	64.8	3.3	5.5	1.9	243	1.5
6c	12.6	19.3	24.9	66.3	3.4	6.5	1.9	316	1.5
6d	11.4	5.9	22.1	58.0	3.3	5.6	2.0	249	2.5
6e	11.3	8.8	22.8	59.5	3.4	5.8	2.0	363	2.1
7a	11.5	15.8	28.7	79.0	3.3	5.3	2.2	186	1.3
7b	11.2	13.4	26.2	47.9	3.4	5.7	2.0	399	2.4
7c	11.4	14.7	32.2	81.8	3.3	5.2	2.2	316	2.5
7d	11.5	18.1	20.7	66.8	3.3	5.3	2.2	441	3.4
7e	11.0	13.3	19.6	57.9	3.4	5.4	2.1	304	2.1
Cloudy FC
1a	11.4	8.2	22.0	65.1	3.4	4.4	2.6	490	3.3
1b	11.2	22.1	35.4	104.1	3.4	4.4	2.5	430	4.6
1c	12.1	7.5	19.7	62.5	3.5	5.4	2.2	387	3.4
1d	11.5	16.5	18.8	60.4	3.2	4.5	2.6	640	6.2
1e	11.5	11.6	20.1	60.8	3.5	4.3	2.7	385	3.4
2a	11.6	20.5	21.3	68.4	3.4	4.9	2.4	557	3.4
2b	11.3	26.6	35.1	60.7	3.4	4.7	2.4	490	5.1
2c	11.6	10.8	26.8	65.3	3.4	5.7	2.0	464	2.8
2d	11.7	11.8	20.7	52.9	3.2	5.0	2.3	757	7.1
2e	11.4	10.9	24.1	63.7	3.4	5.4	2.1	633	6.2
Cloudy NFC
1a	11.5	13.0	25.9	76.3	3.3	5.8	2.0	553	3.1
1b	11.2	16.5	24.6	71.4	3.3	5.4	2.1	630	6.4
1c	12.0	10.8	21.3	60.1	3.4	6.4	1.9	1071	7.1
1d	10.8	18.7	19.1	52.2	3.3	5.1	2.1	761	8.4
1e	11.7	11.7	23.9	62.6	3.4	5.4	2.2	648	9.5
2a	11.4	14.8	20.5	70.1	3.4	6.0	1.9	896	11.9
2b	10.7	14.2	15.8	61.5	3.3	5.5	1.9	1169	17.0
2c	10.9	12.6	17.6	63.1	3.3	5.3	2.0	1925	14.9
2d	10.8	15.9	16.5	54.4	3.3	5.0	2.2	1382	16.4
2e	12.1	21.8	24.2	76.6	3.4	4.9	2.5	1026	9.8
3a	11.7	1.8	23.9	70.8	3.3	6.1	1.9	419	3.2
3b	13.7	5.0	33.4	72.1	3.2	7.8	1.8	647	5.7
3c	14.2	3.8	35.7	78.8	3.2	9.5	1.5	666	5.4
3d	13.6	3.9	32.1	77.2	3.4	8.7	1.6	759	6.8
3e	13.3	5.0	31.3	83.6	3.2	9.1	1.5	619	7.7
4a	11.0	3.3	22.2	69.7	3.6	3.5	3.1	380	2.9
4b	11.5	3.0	28.9	84.9	3.6	3.0	3.8	235	1.9
4c	11.3	4.0	31.7	92.1	3.7	3.2	3.6	183	2.1
4d	12.2	5.5	31.8	86.9	3.6	3.5	3.5	341	3.2
4e	13.6	4.9	35.0	83.5	3.5	4.9	2.8	258	2.4
5a	10.6	0.8	24.9	68.8	3.4	4.3	2.5	393	3.5
5b	11.8	4.0	35.8	90.2	3.3	5.2	2.3	686	4.4
5c	13.5	4.1	32.2	76.9	3.3	8.1	1.7	738	4.8
5d	14.9	6.3	38.1	91.2	3.4	5.0	3.0	717	7.4
5e	12.8	4.3	27.5	67.1	3.5	4.8	2.6	290	3.0
6a	11.2	4.2	24.7	69.1	3.5	4.7	2.4	290	2.2
6b	12.2	5.6	25.4	61.6	3.3	5.5	2.2	171	1.4
6c	13.1	3.1	33.9	60.2	3.5	5.1	2.6	274	1.7
6d	12.8	2.0	29.2	77.7	3.1	6.4	2.0	200	2.1
6e	12.9	5.2	31.4	79.5	3.4	5.0	2.6	196	2.1
7a	11.2	22.8	19.0	70.5	3.5	4.8	2.3	522	5.8
7b	11.2	11.9	33.3	95.7	3.5	3.5	3.2	551	6.0
7c	11.4	23.1	24.4	76.2	3.4	4.3	2.7	567	4.0
7d	11.6	11.3	24.7	71.8	3.3	5.3	2.2	648	6.9
7e	12.2	9.9	21.1	74.6	3.7	4.3	2.8	687	8.4
Fresh
1a	11.4	39.6	20.0	80.4	3.4	4.4	2.6	507	3.4
1b	11.7	19.2	13.6	60.3	3.3	6.1	1.9	278	2.7
1c	10.5	14.7	17.6	57.9	3.5	3.7	2.9	360	0.8
1d	11.4	17.8	15.8	67.9	3.4	4.0	2.9	430	4.9
1e	11.1	27.3	16.6	57.6	3.4	4.6	2.4	481	5.1
2a	10.6	19.8	18.3	59.3	3.4	4.4	2.4	483	2.8
2b	10.9	22.1	14.7	52.1	3.4	4.5	2.4	391	3.0
2c	10.9	22.7	14.5	55.5	3.4	4.2	2.6	415	2.0
2d	10.4	18.7	15.3	57.1	3.5	3.0	3.5	437	5.1
2e	11.3	12.2	11.6	63.0	3.4	4.4	2.5	403	4.6
AIJN	min.	5–30	15–35	45–85	—	2.2–7.5	—	—	—
Code of	10.0/	g/L	g/L	g/L		g CA/L
Practice	min.
limits	11.2*

*Minimum TSS limit for NFC juices: 10.0, for FC juices: 11.2.

An average pH of juices was 3.4 with a small deviation. The TA was on average 5.2 g/L and ranged from 3.0 g/L in fresh juice to 9.5 g/L in naturally cloudy juice. A few juices, all from cloudy NFC category, exceeded the maximum limit of CoP for this parameter. For comparison, the reported values of TA for juices from non-commercial varieties of apples were in range of 2.3–18.2 g/L.^[4] The sugar/acid ratio is a key characteristic determining the flavor of fruit products and in the juices studied was very diverse ranging from 1.5 to 3.8.

The total phenolic contents ranged from 116 to 1925 mg/L and generally agreed with the literature data. Only some samples of one of the cloudy NFC juice exhibited higher values of the total phenolic content and the antioxidant activity than most of those previously reported. These samples of cloudy juice were produced from the Šampion cultivar, as reported on the label. The high content of phenolic compounds in the laboratory-prepared cloudy juice from the Šampion cultivar has been reported, noting that the concentration of the polyphenolic compounds in the apple pulp depends strongly on the apple cultivar and the technology used to produce the pulp.^[8] The high content of total phenolic compounds in the laboratory-prepared juices, exceeding 2000 mg/L was found for some apple cultivars grown in Upper Austria.^[26] On the other hand, added vitamin C (producer declaration) may contribute to the observed high values of the total phenol content and the antioxidant capacity. Ascorbic acid has been reported the major interference in the analysis of wine and most fruit using the Folin–Ciocalteu reagent.^[36]

The differences in the total phenolic compounds content between the juices were reflected in their different total antioxidant capacity quantified by the TEAC value. The high correlation was found between TEAC value and polyphenol content (r = 0.901), showing that these compounds are mainly responsible for the antioxidant capacity of the apple juices. This observation is in agreement with the literature data reported for apples^[37–39] and apple juices.^{[26,27,40,41]}

The chemical composition of apple juices depends on both the apple cultivar and the processing technology. The groups of juices from different product categories were characterized. The descriptive statistical analysis of the chemical properties including mean value, standard deviation, and range for each of the studied categories is presented in Table 2.

Table 2. Chemical characteristics of the apple juices in each of the four categories.

	TSS (°Brix)	Sucrose (g/L)	Glucose (g/L)	Fructose (g/L)	pH	TA (g CA/L)	TSS/TA	Total phenolic (mg GAE/L)	TEAC (mM)
Clear FC	11.4^b ± 0.4	14.3^a ± 6.1	25.1^a ± 5.3	67.0^a ± 11.6	3.3^b ± 0.1	5.4^a ± 0.4	2.1^b ± 0.2	322^b ± 104	2.3^b ± 0.7
n = 35	10.6–12.6	5.9–37.4	16.4–42.4	47.9–87.2	3.1–3.5	4.6–6.5	1.6–2.6	116–602	1.1–4.6
Cloudy FC	11.5^ab ± 0.2	14.6^ab ± 6.4	24.4^a ± 6.2	66.4^a ± 13.9	3.4^ab ± 0.1	4.9^a ± 0.5	2.4^ab ± 0.2	523^ab ± 122	4.6^ab ± 1.5
n = 10	11.2–12.1	7.5–26.5	18.8–35.4	52.9–104.1	3.2–3.5	4.3–5.7	2.0–2.7	385–757	2.8–7.1
Cloudy NFC	12.1^a ± 1.1	8.82^b ± 6.4	26.9^a ± 6.1	73.7^a ± 10.7	3.4^a ± 0.1	5.4^a ± 1.6	2.4^ab ± 0.6	614^a ± 372	6.0^a ± 4.1
n = 35	10.6–14.9	0.8–23.1	15.8–38.1	52.2–95.7	3.1–3.7	3.1–9.5	1.5–3.8	171–1925	1.4–17.0
Fresh	11.0^b ± 0.4	21.4^a ± 7.6	15.8^b ± 2.4	61.1^a ± 8.0	3.4^ab ± 0.1	4.3^a ± 0.8	2.6^a ± 0.4	418^ab ± 67	3.4^ab ± 1.4
n = 10	10.4–11.7	12.2–39.6	11.6–20.0	52.1–80.4	3.3–3.5	3.0–6.1	1.9–3.5	278–507	0.8–5.1

TSS: total soluble solids, TA: titratable acidity, TEAC: Trolox equivalent antioxidant capacity.

Different superscripts in the same column indicate significant difference between samples (p ≤ 0.05), determined by Tukey’s test.

The values of the chemical parameters varied significantly within each group of juices. The cloudy NFC juices present the most diverse chemical properties. On the other hand, when the mean values were considered, the groups of juices exhibited between-group variability in the parameters under study. Highly significant differences (p < 0.05) were found between the mean values of most of the chemical parameters for every class of the juices studied, Table 2.

The highest TSS content was measured for cloudy NFC juices, the mean value was significantly higher than for clear FC and fresh juices. The composition of the individual sugars also varied between the product classes. Fresh and clear FC juices had the highest content of sucrose, cloudy FC juices had intermediate values, and cloudy NFC juices the lowest. The concentration of glucose in clear FC, cloudy FC and cloudy NFC juices was similar and significantly higher than that in fresh juices. The content of fructose did not differ significantly among studied juices.

All juice classes showed similar TA values and some differences in pH values. The highest mean values of total soluble solids to titratable acidity (TSS/TA) ratio were found in fresh varieties, and the lowest in clear FC juices. The naturally cloudy NFC juices had significantly higher content of polyphenols then clear FC juices. Cloudy FC and fresh juices had intermediate content of polyphenols. Clear FC juices had the lowest content of polyphenols due to the specifics of the production process.

Exploratory analysis of chemical profiles of apple juices

The analysis of the individual chemical parameters revealed diversity of chemical properties among the juices from different categories. Multivariate unsupervised pattern recognition methods were applied for the detailed comparison of the overall chemical profiles of the particular apple juices.

PCA

First, the chemical data were analyzed by the PCA to perform the characterization of the juices. The matrix of nine chemical variables for the 90 juices was analyzed. The plots of the scores for the PCA of the data are shown in Fig. 1, and the correlation loadings for the first four PCs are presented in Table 3.

Table 3. Loadings for the first four principal components for apple juice samples.

	PC1	PC2	PC3	PC4
TSS	0.479	0.675*	–0.158	–0.177
Sucrose	–0.160	–0.529	0.104	0.754*
Glucose	0.182	0.857*	–0.113	0.275
Fructose	–0.070	0.731*	–0.326	0.473
pH	–0.667*	0.295	–0.394	–0.126
TA	0.922*	0.189	0.191	–0.0004
TSS/TA	–0.849*	0.167	–0.361	–0.114
Total phenolic	0.454	–0.413	–0.755*	0.029
TEAC	0.393	–0.409	–0.794*	–0.013

*Significant loadings.

TSS: total soluble solids, TA: titratable acidity, TEAC: Trolox equivalent antioxidant capacity.

Figure 1. Principal component analysis of the chemical profiles of 90 apple juices, scores plots A: PC1 versus PC2; and B: PC3 versus PC4.

The first four principal components (PC1, PC2, PC3, and PC4) accounted, respectively, for 30, 28, 19, and 10% of the variance of the experimental data. The projection of the sample points onto the plane defined by the first two principal components PC1 and PC2 revealed only some differences of the fresh juices, which were to some extent separated from the other samples. Both PC1 and PC2 contributed to this separation. The largest dispersion in the plane defined by PC1 and PC2 was observed for the cloudy NFC juices, illustrating that these latter juices have the most variable chemical profiles.

The first PC was positively correlated to TA, and negatively correlated to TSS/TA ratio and pH value. The second PC was positively correlated to glucose, fructose and TSS. In contrast, the scores plot in the plane defined by PC3 versus PC4 showed the differentiation of the chemical profiles between clear FC juices and cloudy NFC juices. These two groups were separated along the PC3 axis. This component was negatively correlated with the total polyphenols content and total antioxidant capacity. As apparent from the sample distribution, the cloudy NFC juices were generally richer in polyphenols and had higher antioxidant capacity as compared to clear FC juices. The juices of the remaining two groups—cloudy FC and fresh juices (show intermediate antioxidant properties)—were not separated in this plane.

Cluster analysis

Cluster analysis was another unsupervised pattern recognition technique used for preliminary evaluation of the multivariate chemical profiles of the juices. HCA using Ward’s method was applied to pre-processed chemical variables. The matrix of nine chemical variables was analyzed for the 90 juices. Using this technique, samples were grouped on the basis of similarities in their chemical composition, Fig. 2.

Figure 2. Hierarchical cluster analysis of the chemical profiles of 90 apple juices. Juices: clear FC (red), cloudy FC (green), cloudy NFC (blue), fresh (cyan).

Five sample clusters were detected, with the first containing clear FC juices (the majority of the studied samples). The second cluster contained all of the studied fresh juices and single juices from other classes. The cloudy NFC juices formed three distinct classes, confirming their chemical diversity. Cloudy FC juices did not form separate cluster and were scattered over all of the remaining clusters. Both of the applied unsupervised techniques, PCA and HCA, revealed clustering of juices that to some extent reproduces the juice categories. Clear FC and cloudy NFC juices had the most distinct chemical profiles.

Classification of the apple juices

In order to test the possibility of classification of juices based on their chemical properties we used kNN classification method. In this method the information about the class membership of juices in their respective category was used to classify new juice samples into one of the known categories based on their chemical properties. The details of the classification model developed using the calibration set are shown in Table 4. The model obtained was evaluated in terms of sensitivity, specificity, and misclassification error.

Table 4. Characteristic of the kNN classification model.

	Sensitivity (%)	Specificity (%)	Classification error (%)
Clear FC	91	76	17
Cloudy FC	0	98	51
Cloudy NFC	81	85	17
Fresh	50	100	25

The sensitivity of a class model is defined as the rate of objects belonging to the class which is correctly identified by the mathematical model.^[11] The highest sensitivity was obtained for the clear FC and cloudy NFC classes. The sensitivity for the fresh juice class was considerably lower. The sensitivity was zero for the cloudy FC juices, meaning that none of the samples of this class was classified correctly. Specificity was determined as the rate of objects foreign to the class that are classified as foreign.^[11] High specificity was determined for all of the classes studied. Classification errors were relatively low for the clear FC and cloudy NFC juices and a little higher for fresh juices. High classification error of 51% was obtained for the class of cloudy FC juices.

The developed classification model was used for classification of the samples from the test set, Table 5. As one can expect, correct classification was obtained for clear FC juices—all of the samples were correctly classified into the respective category. Good classification results were also obtained for cloudy NFC and fresh juices. The model failed to classify cloudy FC juices—all of the samples were incorrectly assigned to either the clear FC or cloudy NFC categories. The poor classification of these juices might result from the similarity of their chemical profiles to the samples of other categories. The number of samples in the cloudy FC and fresh categories was much lower than those of the two other categories, which may have affected the classification analysis. In fact the classification model showed inferior performance for the fresh category as compared to the clear FC and cloudy NFC juices. However, fresh juices have chemical profiles distinct from those of the other juices, and, therefore, the results for this category were better than for the cloudy FC. When all of the test samples were considered, 77% of juices were classified correctly.

Table 5. Prediction results for the test set samples using kNN classification model.

Actual	Clear FC	Cloudy FC	Cloudy NFC	Fresh
Predicted as
Clear FC	13	4	2	0
Cloudy FC	0	0	0	0
Cloudy NFC	0	1	6	0
Fresh	0	0	0	4

Conclusion

The commercial apple juices studied in this work showed a wide variability in their chemical characteristics. Multivariate methods were particularly useful for the data analysis, as the quality of juices was assessed on the basis of several parameters. Exploratory study revealed significant differences among the chemical profiles of the commercial apple juices from different categories, on the other hand, similarity of juices within a given category was observed. The cloudy FC juices were an exception, as they did not form a separate group according to their chemical profiles. The classification results seem to be encouraging, especially when we take into account the multiple sources of variability of the chemical composition of apple juices: diversity of the raw material characteristics, details of the production technology, and the effects of storage conditions. The results indicate that the chemical profiles are, to some extent, characteristic for the specific juice category. Altogether, the results indicate that it should be possible to develop methods to identify the category of juices based on their chemical characteristics. Such models could be useful in traceability and authentication of apple juices. However, the development of representative and robust predictive models for real-world applications would require an increase in the number and diversity of samples and studies of expanded chemical profiles.

Conflict of interest

The authors declare no conflict of interest.