Skip to main content
NCBI home page
Food Chemistry: X logoLink to Food Chemistry: X
. 2025 Sep 25;31:103082. doi: 10.1016/j.fochx.2025.103082

In-depth analysis of the characteristics of volatile organic compounds in wines: a systematic study integrating intelligent sensory and metabolomics techniques with chemometrics and machine learning models

Rui Xie a, Jiawen Liu a, Yutao Li a, Yong Chen b, Tian Shen c, Meilong Xu c, Yanlun Ju a, Yulin Fang a,⁎, Zhenwen Zhang a,⁎
PMCID: PMC12513181  PMID: 41080139

Abstract

The volatile organic compounds (VOCs) in wines of ‘Dornfelder’ (DF), ‘Petit Verdot’ (PV), ‘Pinot Noir’ (PN), ‘Sangiovese’ (SV) and ‘Malbec’ (MB) were analyzed using an E-nose, HS-SPME-GC–MS and HS-GC-IMS. A total of 94 VOCs were identified by two techniques. Specifically, HS-SPME-GC–MS identified 70 compounds (alcohols' concentration accounting for 52.56%–68.75 %), and HS-GC-IMS identified 36 compounds (esters' concentration accounting for 35.58 %–42.05 %), with 12 compounds were identified by both methods. 15 key differential VOCs identified through chemometrics and machine learning analysis. Additionally, correlation analysis of E-nose sensor responses with key differential VOCs indicated that W2S, W2W, and W5S may be more suitable for predicting levels of 2-methylbutyl acetate, 3-methyl-butanoic acid, and isoamyl acetate, which can thus help to quickly identify PV wine. These results help to understand the flavor differences between different varieties of wines and provide a theoretical basis for wine flavor differentiation, quality control and product development.

Keywords: Wine, E-nose, HS-SPME-GC–MS, HS-GC-IMS, Volatile organic compounds, Chemometrics, Machine learning

Chemical compounds studied in this article: Phenylethyl alcohol (PubChem CID: 6054), (Z)-2-Hexen-1-ol (PubChem CID: 5324489), 3-Methyl-1-butanol (PubChem CID: 31260), 2-Methyl-1-propanol (PubChem CID: 6560), Ethyl acetate (PubChem CID: 8857), 2-Methylbutyl acetate (PubChem CID: 12209), Ethyl hexanoate (PubChem CID: 31265), Isoamyl acetate (PubChem CID: 31276), 3-Methyl-butanoic acid (PubChem CID: 10430), Terpinen-4-ol (PubChem CID: 11230)

Highlights

  • •

    E-nose, HS-SPME-GC–MS and HS-GC-IMS were employed to analyze VOCs in five wines.

  • •

    HS-SPME-GC–MS can sensitively identify alcohols, while HS-GC-IMS shows higher sensitivity in esters.

  • •

    NB, RF, KNN and DT algorithms are used to construct models, and RF model shows extremely excellent performance.

  • •

    Identified 15 key differential VOCs by chemometrics and machine learning modeling analysis.

  • •

    Established the correlation between E-nose sensors and key differential VOCs, which helps to quickly identify PV wine from others.

1. Introduction

Wine is an ancient and popular low-alcohol beverage (Álvarez-Fernández et al., 2020), rich in nutrients such as sugars, vitamins, amino acids, proteins, minerals, polyphenols, volatiles, etc., whose interactions and transformations give wine a unique organoleptic experience and greater nutritional value (Fia et al., 2018). Many studies have shown that moderate wine consumption is beneficial to human health, and plays an important role in protecting the skin and preventing neurological and cardiovascular diseases (Vecchio et al., 2017). The sensory quality of wine is mainly considered in terms of balance, intensity, complexity, and finish, and volatile organic compounds (VOCs) are important parameters that influence the sensory characteristics of wine (Jiang & Zhang, 2018). Therefore, the identification and analysis of VOCs in wine is important.

In recent years, with the continuous advancement of analytical technologies, methods such as the electronic nose (E-nose), headspace solid-phase microextraction gas chromatography–mass spectrometry (HS-SPME-GC–MS), and headspace gas chromatography-ion mobility spectrometry (HS-GC-IMS) have found extensive applications in the field of VOC analysis. The E-nose mimics the human olfactory system. It consists of a set of partially specific chemical sensors and detection analysis software, capable of providing comprehensive information on the flavor of samples. This technology offers several advantages, including convenient measurement, rapidity, high sensitivity, low cost, and non-destructiveness (Liu et al., 2022). HS-SPME-GC–MS has been extremely widely used in the identification and quantitative analysis of food flavor components. It features a broad detection range and comprehensive fragment information in the NIST spectral library, excelling particularly in the qualitative analysis of unknown substances. However, samples usually require enrichment and concentration, which may lead to changes in their components (Yan et al., 2021). As an emerging technology for flavor analysis, HS-GC-IMS focuses on the detection of trace VOCs (with molecular weights and boiling points less than 300 Da/°C). It exhibits high sensitivity, eliminating the need for sample enrichment and concentration, thereby preserving the true flavor of samples to the greatest extent. In addition, this technology is easy to operate and offers high-level data visualization (Feng et al., 2022). It has been successfully applied to the identification of various commercial essential oils, such as olive oil (Gerhardt et al., 2019), celery seed oil (Xu et al., 2023), walnut oil (Xi et al., 2024), and personal care products like cosmetics (Rodríguez-Maecker et al., 2017). Nevertheless, due to the incomplete GC-IMS database, it remains challenging to identify some signal peaks in the spectra (Wei et al., 2023).

Meanwhile, food flavor analysis has entered the 4.0 era. Compared with traditional classification techniques, machine learning (ML) demonstrates remarkable advantages in identifying commonalities and differentiating sample characteristics (Zeng et al., 2023). Many studies have integrated ML with food flavor analysis, applying it to the screening of characteristic compounds and food identification (Li et al., 2024; Wang et al., 2025). In the field of wine, the aroma profile of wine as crucial indicators of its quality. Moreover, the differences in aroma attributes and the composition of VOCs among wines of different varieties provide essential references for quality control, product development, and market positioning within the wine industry.

Given the distinct characteristics of each technique, it is highly necessary to combine multiple methods for evaluating VOCs in wines. In this study, we utilized analytical techniques such as the E-nose for intelligent sensory evaluation, HS-SPME-GC–MS, and HS-GC-IMS to obtain information on the VOCs in wines of five grape varieties. We then employed principal component analysis (PCA), orthogonal partial least squares-discriminant analysis (OPLS-DA), as well as naive bayes (NB), random forest (RF), K-nearest neighbors (KNN), and decision tree (DT) models to analyze VOCs. Additionally, odor activity value (OAV), relative OAV (ROAV) and variable importance in projection (VIP) values were used to evaluate the key differential compounds. This research will provide novel perspectives and methods for a deeper understanding of the aroma characteristics and quality evaluation of wines from different grape varieties.

2. Materials and methods

2.1. Samples and reagents

2.1.1. Wine samples

Five grape varieties were used in this study: ‘Dornfelder’ (V. vinifera L., DF), ‘Petit Verdot’ (V. vinifera L., PV), ‘Pinot Noir’ (V. vinifera L., PN), ‘Sangiovese’ (V. vinifera L., SV), and ‘Malbec’ (V. vinifera L., MB). These varieties were obtained from Chateau Greatwall Terroir in Ningxia, China (37°43′–39°23′ N, 105°45′–106°47′ E). The vineyard was planted with 7-year-old vines that were own-rooted with 3.5 m and 1.0 m inter- and intra-row spacing, respectively. Modified Vertical-Sloping-Positioning (M-VSP) trellis system, hedgerow-type leaf screen, and routine management was performed according to local standards. Harvested in late August and early September 2023 (°Brix >23 %). Wines were vinified according to the procedure described in study by Sun et al. (2018). Wines were stored at 4 °C before analysis. The basic information of wines is shown in Table 1.

Table 1.

Basic information about the five wines.

Varieties Alcohol by volume (v/v) Volatile acid (g/L) Titratable acid (g/L) pH
DF 12.38 ± 0.01e 0.41 ± 0.02c 5.84 ± 0.01b 3.67 ± 0.00b
PV 13.71 ± 0.05c 0.61 ± 0.04b 5.88 ± 0.03a 3.63 ± 0.00c
PN 12.88 ± 0.05d 0.65 ± 0.01b 5.56 ± 0.01d 3.71 ± 0.00a
SV 14.95 ± 0.01a 0.81 ± 0.03a 5.52 ± 0.01d 3.67 ± 0.02b
MB 14.67 ± 0.01b 0.61 ± 0.04b 5.79 ± 0.01c 3.67 ± 0.00b
Open in a new tab

Different letters in the same column indicate significant differences (p < 0.05).

2.1.2. Chemical reagents

Ethanol (chromatographic grade, purity ≥99.7 %, Lot No.: 20240210) was purchased from Kemiou Chemical Reagent Co., Ltd. (Tianjin, China). The internal standard (4-methyl-2-pentanol, chromatographic grade, purity ≥99 %, Lot No.: C11619742) was purchased from Sinopharm Chemical Reagent Co., Ltd. (Beijing, China). Reference standards of aroma compounds (including alcohols, esters, acids, ketones, phenols, aldehydes, and terpenes) and n-ketones (C4–C9) (chromatographic grade, purity ≥99 %) were purchased from Sigma-Aldrich Chemical Reagent Co., Ltd. (Shanghai, China).

2.2. E-nose analysis

E-nose (Airsense PEN3-Plus, Germany) was used for the initial detection and analysis of aroma profiles for five grape variety wines. The device has 10 metal oxide sensors, W1C (aromatic), W5S (nitrogen oxides), W3C (ammonia, aromatic), W6S (hydrogen), W5C (arom-aliph), W1S (broad-methane), W1W (sulphur-organic), W2S (broad-alcohol), W2W (sulph-chlor), W3S (methane-aliph), which are sensitive to different groups of VOCs, as well described in previous studies (Chen et al., 2020).

5 mL of the sample was added to 15 mL injection vial, sealed with parafilm and allowed to stand for 30 min at room temperature. The injection flow rate was 400 mL/min, the preparation time was 5 s before starting measurement, and the measurement lasted 100 s. The computer recorded signal data from sensor once per second. After each measurement, zero gas (activated carbon-filtered air) was pumped into the sample gas line to purge sensor. The measurement data was processed using Win Muster software. Samples were analyzed in triplicate.

2.3. HS-SPME-GC–MS analysis

VOCs were detected using the 7890B GC-5977B MS (Agilent, USA) in combination with HS-SPME. Detection and analysis methods were as described in Yang et al. (2023). Chromatographic column was HP-INNOWax (60 m × 250 μm × 0.25 μm). Carrier gas was hexane (purity >99.999 %, flow rate 1 mL/min). The inlet temperature was 250 °C, split injection mode of 5:1 (flow rate of 5 mL/min) was used, the injection volume was 0.5 μL; heating procedure: 50 °C was held for 1 min, and then increased to 220 °C at a rate of 3 °C/min for 10 min. The mass spectrometer was operated in electron ionization (EI) mode at 70 eV, ion source temperature at 230 °C, and the mass spectral interface temperature at 280 °C. Detection was performed in SCAN all-ion scan mode (range of 43–450 u). Samples were analyzed in triplicate. Compounds in the GC–MS full-scan data were identified using an automated deconvolution technique based on chromatographic retention index (RI), mass spectra, and the results of NIST14.L standard spectral libraries of existing standards. VOCs with corresponding standards were quantified directly from the standard curves of their standards, compounds without standards were analyzed semi-quantitatively according to principles of chemical structure similarity and proximity in carbon atom count.

2.4. HS-GC-IMS analysis

VOCs were detected by HS-GC-IMS (Flavor Spec®-G.A.S., Dortmund, Germany). Sample analyses were determined with reference to method of Yin et al. (2023) with some practical modifications. 5 mL of sample was added to 15 mL headspace vial and incubated for 15 min at 40 °C, 500 rpm with shock heating. Chromatographic column was an MXT-5 (15 m × 0.53 mm, 1.0 μm), the automatic headspace injection volume was 100 μL, and volatiles were separated in positive ion mode. Samples were analyzed in triplicate. Standard curves were constructed using n-ketones (C4-C9) as an external reference, and volatiles were identified by comparing RI and drift time of compounds in NIST RI database and ion mobility spectrometry library. VOCs were quantified based on HS-GC-IMS peak intensity (peak volume). The total analysis time for samples was 30 min, and the specific program settings are detailed in Table S1.

2.5. Calculation of OAV and ROAV

The formula of OAV and ROAV is:

OAVi=CiTi;ROAVi=CiTi×TmaxCmax×100

where Ci and Ti represent the concentration of compound i and the odor threshold of this compound found in the literature, respectively. Tmax/Cmax is the maximum of Ti/Ci among all the compounds in the sample (Wang et al., 2020).

2.6. Statistical analysis

All experimental data were presented as mean ± standard deviation in triplicate. Data collected between 64 and 70 s were chosen for analysis, as the E-nose sensor exhibited greater stability after 58 s and before 75 s. A one-way analysis of variance (ANOVA) was conducted using PASW Statistics 18 (IBM, Armonk, NY, USA), and post-hoc Duncan's test with a significance level of p < 0.05 revealed significant difference. Heat map was generated using TB tools (Cjchen, China). OPLS-DA was carried out with SIMCA 14.1 (MKS Umetrics AB, Umea, Sweden). GC-IMS data were processed using Laboratory Analytical Viewer processing software. Origin 2021b was used to perform PCA and Spearman correlation analysis, and to generate radar plots and pie charts (OriginLab, Northampton, MA, USA). In Spearman's correlation analysis, p < 0.05 indicates a significant correlation, and p < 0.01 indicates an extremely significant correlation. The ML models, NB, RF, KNN and DT were constructed using Python language, and the performance of models was evaluated using receiver operating characteristic (ROC) curves as well as the area under curve (AUC) values.

3. Results and discussion

3.1. E-nose aroma profiles

E-nose has been widely used in aroma recognition and discrimination of products due to its capability to objectively acquire the aroma profiles of VOCs (Zhu et al., 2017). In this study, radar chart (Fig. 1A) shows that 5 sensors (W1S, W1W, W2S, W2W and W5S sensors) have high response values, which means that corresponding aroma could be considered as characteristic differential aroma responsible for discriminating five wines. It is noteworthy that W1S sensor exhibits the highest response value among all samples, which is sensitive to methyl group of substances (Chen et al., 2020). PCA can reduce the dimensionality of multidimensional data matrices while preserving overall information of original data and linear classification of feature vectors (Baldwin et al., 2011). As shown in Fig. 1B, the cumulative variance explained by PC1 and PC2 was 86.18 % (74.51 % and 11.67 % for PC1 and PC2, respectively), indicating that two principal components can effectively reflect main aroma characteristics of different samples. The value of PC1 decreases in order of PV, DF, SV, MB, PN and the value of PC2 decreases in order of SV, MB, PN, PV, DF. By combining the spatial distribution of samples in the PCA plot, we observed a significant similarity in aroma attributes between SV and MB. In addition, 3 sensors (W1S, W2S and W5S) are located near PV, 2 sensors (W3C and W5C) are close to PN, the W2W and W1C sensors are relatively close to SV and MB, and DF is much closer to W6S. These characteristics mean that compared with other samples, PV showed higher correlations with methyl, alcohol, aldehyde-ketone and nitrogen oxide aromas. PN exhibited stronger correlations with short-chain alkanes and benzenes. SV and MB showed stronger inorganic sulfide and aromatic aromas. DF showed stronger hydride aromas. Although E-nose can effectively identify the overall aroma profiles of different samples, it cannot provide detailed information on specific VOCs. Therefore, we subsequently employed HS-SPME-GC–MS and HS-GC-IMS techniques to identify and conduct more in-depth analysis of VOCs.

Fig. 1.

Fig. 1

Open in a new tab

Radar chart (A) and PCA double plots (B) of aroma profiles in five grape variety wines recorded by E-nose.

3.2. VOCs detected by HS-SPME-GC–MS

This study identified 70 VOCs that can be categorized into 8 categories, including alcohols (13), acids (7), C6/C9 compounds (17), esters (16), terpenes and norisoprenoids (10), ketones (1), volatile phenols (2) and aldehydes (4) (Table S2). Among them, alcohols' concentration accounting for the largest proportion of the total number of identified compounds in wines (52.56 %–68.75 %), followed by acids (14.76 %–23.44 %) (Fig. 2A), which is similar to the findings of Yang et al. (2023). Alcohols are mostly associated with floral, herbal, balsamic and chemical odors (Kong et al., 2019); short-chain fatty acids are associated with negative odors and they tend to have a higher threshold, with cheese and cream odors at low concentrations (Shinohara, 1985). C6 compounds are green-odor compounds, higher ratios of C6 to C9 compounds are associated with more pronounced green aromas (Escudero et al., 2007). As important odorants in wine, esters have intense tropical fruit and pleasant floral aromas (Zhang et al., 2013). Terpenes and norisoprenoids are commonly associated with floral, sweet fruit and lemon aromas (Zhao et al., 2019). Only one ketone (3-hydroxy-2-butanone) was detected in this study, which may contribute to buttery, creamy flavors (Peinado et al., 2004). Volatile phenols are often described as balsamic and chemical odors. Aldehydes can impart fruity and fresh vegetal aromas to wine (Li et al., 2021).

Fig. 2.

Fig. 2

Open in a new tab

Proportion of each category of VOCs in five wines detected by HS-SPME-GC–MS (A) and HS-GC-IMS (B). Numbers 1–5 show the percentage of each category of VOCs in samples DF, PV, PN, SV and MB, respectively.

The total concentration of VOCs detected by HS-SPME-GC–MS followed the order SV > PV > MB > DF > PN. Among these, SV had the highest concentrations of alcohols, acids, and aldehydes, but the lowest concentrations of C6/C9 compounds and ketones; PV had the lowest concentration of esters; MB had the lowest concentration of terpenes and norisoprenoids and volatile phenols; DF had the highest concentration of C6/C9 compounds, esters and volatile phenols; PN had the highest concentration of terpenes and norisoprenoids, ketones and the lowest concentration of acids. These characteristics explain the significant differences in aroma attributes exhibited by different varieties of wines.

3.3. VOCs detected by HS-GC-IMS

The VOCs of five wines were identified by HS-GC-IMS. Qualitative results are shown in Fig. S1, where the X-axis represents drift time and the Y-axis represents RI. Red numbers in the graph indicate identified organic compounds (detailed information is provided in Table S3), and signals of most compounds appear in the RI range of 200–800 s and drift time range of 1.0–2.0 s. HS-GC-IMS identified 36 compounds, including alcohols (5), esters (10), aldehydes (3), ketones (3), terpenes (3), acids (1), nitrogen heterocycles (2) and C6 compounds (9) (Table S3), with the highest percentage of esters (35.58 %–42.05 %), followed by C6 compounds (20.62 %–25.68 %) (Fig. 2B).

Characteristic fingerprints obtained by Gallery Plot provide a more intuitive indication of differences in VOCs from different wines. Each row represents the signal intensity of different compounds in the same sample, and each column represents that of the same compound in different samples (the deeper the red color, the larger the concentration) (Xi et al., 2024). Detailed information on signal peak intensities is shown in Table S4. As shown in Fig. 3A, the signal intensity of 3-methyl-3-butenol and hexyl acetate were much higher in DF than others. The maximum signal intensity of hexanal was found in PV. High levels of 2,3-butanedione and propanal were found in PN. SV showed the significantly highest levels of 2-methyl-1-propanol, ethyl isovalerate and ethyl 2-methylbutyrate. High levels of 2-hexanone and ethyl butyrate were found in MB. Overall, there were significant differences (p < 0.05) in the concentration of VOCs between five wines. The concentration of VOCs in PN and DF were significantly lower than others, which was consistent with GC–MS results.

Fig. 3.

Fig. 3

Open in a new tab

Analysis results of VOCs in five grape variety wines detected by HS-GC-IMS. Map of characteristic fingerprints (A). Two-dimensional top view (B). Diagram comparing the differences (C). Numbers 1–5 represent samples DF, PV, PN, SV, MB, respectively.

For visualization, we used Reporter Plot to create 2D top view of different samples (Fig. 3B). Different colors represent different concentrations, with deeper red color means higher concentration. Select the spectrum of DF as reference for other samples to create comparison graph (Fig. 3C). The red part indicates that concentration of compound is higher than the reference value, and the blue part indicates that concentration of compound is lower than reference value (the deeper color means greater difference in concentration of VOCs). From this, we can clearly see the differences between samples. It has been reported that HS-GC-IMS can be used for identification and differentiation of food species, such as celery (Xu et al., 2023) and goji berries (Zhou et al., 2023). Chen et al. (2020) and Xiao et al. (2022) analyzed variations in VOCs during the ripening of finger citron (Citrus medica L. var. sarcodactylis) and the fermentation of black tea, respectively. Cao et al. (2022) applied this method to analyze VOCs of nine different varieties wine. The results of this study showed that HS-GC-IMS was also able to better distinguish five grape variety wines.

3.4. Multivariate statistical analysis of VOCs

A total of 94 VOCs were identified by HS-SPME-GC–MS and HS-GC-IMS, with 12 detected by both methods, 58 detected only by HS-SPME-GC–MS, and only 24 detected by HS-GC-IMS. Comparing the results of these two methods, more VOCs were detected by HS-SPME-GC–MS, encompassing alcohols, acids, esters, C6/C9 compounds and terpenes. However, GC-IMS was proved to be more sensitive in identifying esters, such as propyl acetate, ethyl isovalerate, methyl 2-methylbutanoate, methyl acetate, 2-methylbutyl acetate, ethyl propanoate etc. Thus, integration of these two methods can lead to more comprehensive investigation of VOCs. Xi et al. (2024) also reached similar conclusions in their study on walnut oil.

OPLS-DA is a supervised statistical method for discriminant analysis that allows further quantification of the extent to which VOCs contribute to the flavor of samples based on VIP value in model (Jin et al., 2021). R2 > 0.5 and Q2 > 0.5, and it indicates good discriminant analysis ability (Yun et al., 2021), which means that the model is reliable (Li et al., 2022; Stranska et al., 2021). In this experiment, OPLS-DA was performed with five wines as X variables and the concentration of VOCs as Y variables (Fig. 4). The results showed better differentiation of different samples in HS-SPME-GC–MS model (R2X = 1.000; R2Y = 1.00; Q2 = 1.000), and the cross-validation after 200 substitution tests showed that the model validation was effective (Q2 = −1.39), and same results were obtained in GC-IMS model (R2X = 0.996; R2Y = 0.998; Q2 = 0.971), and in post-replacement test Q2 = −1.23 (Fig. 4, A2 & B2). DF and PN located in the positive direction of X-axis, PV, SV and MB located in the opposite direction of X-axis and PV and MB are closer in space (Fig. 4, A1 & B1). It is obvious that there is a tendency of clustering for DF & PN, PV & MB and SV. To further clarify the flavor differences among five wines, cluster analysis was conducted using the 106 VOCs as variables. Analysis based on the HS-SPME-GC–MS data (Fig. 4, A3) showed that five wines could be classified into three categories, namely SV, PV & MB and PN & DF, and SV was closer to PV and MB. The analysis based on HS-GC-IMS data (Fig. 4, B3) indicated that the samples were classified into four categories. PV & MB formed one category, and the others were each in a separate category. Moreover, DF tended to be grouped with PN, and SV tended to be grouped with PV & MB. These results were consistent with those of the OPLS-DA analysis.

Fig. 4.

Fig. 4

Open in a new tab

Multivariate statistical analysis of HS-SPME-GC–MS (A) and HS-GC-IMS (B). OPLS-DA score plot (A1, B1). Cross-validation plot (A2, B2). Heatmap of cluster analysis results (A3, B3).

Aroma quality of wine depends on the combined effect of several compounds, some of which, although high in concentration, are not easily perceived by human olfaction due to their high threshold values (Chen et al., 2021). In general, OAV > 1 or ROAV ≥1 are identified as key aroma compounds responsible for sample (Noguerol-Pato et al., 2012), and VIP > 1 is considered to account for differences between groups (Jin et al., 2021). The OAV/ROAV, VIP value were calculated of VOCs as shown in Table 2. Combining OAV > 1 or ROAV ≥1 and VIP > 1, 8 VOCs were regarded as key differential compounds, with methyl analogs accounting for the largest proportion, which is consistent with the results of E-nose analysis (with higher levels of W1S sensor). Among them, phenylethyl alcohol, (Z)-2-hexen-1-ol, (E)-2-hexen-1-ol, ethyl acetate and terpinen-4-ol contributed significantly for all samples. In addition, 3-methyl-1-butanol contributed significantly to the flavor of DF, PV, SV and MB, 2-methyl-1-propanol to PV, SV and MB, and 2-methylbutyl acetate to DF and MB. The compounds with the highest VIP values were 3-methyl-1-butanol (4.5) followed by phenylethyl alcohol (3.45), and these two compounds had the highest OAV of 11.35 and 25.83 in SV, respectively. Therefore, combined with odor descriptions, we concluded that SV exhibits more perceivable flowery (rose-like) and spicy (whisky) characteristics compared to others.

Table 2.

The OAV/ROAV values and VIP predictions from OPLS-DA of volatile organic compounds in five wines.

Identification methods Number Compounds OAV/ROAV
 VIP pred
DF PV PN SV MB
HS-SPME-GC–MS 1 Phenylethyl alcohol 16.86 20.64 10.17 25.83 15.86 3.45
2 1-Propanol 0.07 0.03 0.10 0.04 0.03 2.31
3 3-Methyl-1-butanol 1.24 9.97 0.72 11.35 9.87 4.50
4 2-Methyl-1-propanol 0.64 1.76 0.68 1.63 1.98 2.10
5 1-Pentanol 0.08 0.02 0.00 0.00 0.02 1.31
6 1-Butanol 0.01 0.01 0.01 0.01 0.01 0.17
7 3-(Methylthio)-1-propanol 1.21 1.55 0.85 2.29 1.34 0.25
8 Benzyl alcohol 0.00 0.00 0.00 0.00 0.00 0.24
9 1-Octanol 0.02 0.01 0.02 0.01 0.01 0.04
10 2-Ethyl-1-hexanol 0.01 0.01 0.01 0.01 0.01 0.02
11 1-Heptanol 0.01 0.01 0.10 0.12 0.12 0.14
12 1-Decanol 0.00 0.01 0.01 0.01 0.01 0.03
13 1-Octen-3-ol 0.00 0.00 0.00 0.00 0.00 0.00
14 Acetic acid 0.61 0.94 0.26 0.98 0.87 3.12
15 Propanoic acid – – – – – 0.56
16 Decanoic acid 0.11 0.07 0.05 0.04 0.05 0.43
17 3-Methyl-butanoic acid 34.49 40.12 18.09 39.06 38.48 0.29
18 2-Methyl-propanoic acid 0.00 0.00 0.01 0.00 0.01 0.82
19 Octanoic acid 1.68 1.01 1.14 0.64 0.88 0.25
20 Benzoic acid – – – – – 0.12
21 (Z)-2-Hexen-1-ol 91.16 56.95 59.18 45.53 48.70 1.57
22 (E)-2-Hexen-1-ol 78.44 48.99 50.93 39.18 41.91 1.46
23 1-Hexanol 0.17 0.10 0.17 0.07 0.17 0.46
Identification methods Number Compounds OAV/ROAV
 VIP pred
DF PV PN SV MB
HS-SPME-GC–MS 24 3-Methyl-1-pentanol 0.46 5.08 8.30 0.41 8.12 0.83
25 4-Methyl-1-pentanol 0.00 0.00 0.00 0.00 0.00 0.17
26 (E)-3-Hexen-1-ol 0.11 0.66 0.08 0.26 0.10 0.46
27 (Z)-3-Hexen-1-ol 0.07 0.30 0.04 0.25 0.10 0.22
28 3-Hydroxybutyric acid ethylester 0.01 0.00 0.01 0.00 0.00 0.13
29 Isobutyl acetate 0.05 0.03 0.04 0.03 0.04 0.10
30 Ethyl butyrate 0.51 0.22 0.79 0.31 0.24 0.03
31 Methyl caprylate 0.04 0.02 0.01 0.01 0.01 0.03
32 Whiskey lactone 0.05 0.10 n.d. 0.11 n.d. 0.05
33 Ethyl isobutyrate 0.08 0.15 0.05 0.19 0.15 0.01
34 Ethyl salicylate – – – – – 0.01
35 Hexanoic acid 0.36 0.21 0.23 0.17 0.22 0.32
36 Nonanal 0.12 0.19 0.07 0.07 0.09 0.01
37 2-Methoxy-3-isobutyl pyrazine 6.82 1.34 6.89 0.98 1.02 0.00
38 Ethyl acetate 5.06 4.47 4.93 5.35 4.78 1.90
39 Isoamyl acetate 143.35 146.78 19.61 80.73 114.55 0.99
40 Ethyl caprylate 5.48 3.44 3.58 2.75 2.94 0.46
41 Ethyl caprate 9.66 5.29 6.03 3.98 4.10 0.37
42 Ethyl lactate 0.01 0.01 0.01 0.01 0.01 0.51
43 Ethyl hexanoate 16.65 9.03 9.62 7.63 9.37 0.37
44 Phenethyl acetate 1.40 1.02 0.10 0.39 0.70 0.29
45 Hexyl acetate 0.12 0.03 0.00 0.01 0.03 0.21
46 Ethyl laurate 0.25 0.09 0.13 0.05 0.06 0.11
Identification methods Number Compounds OAV/ROAV
 VIP pred
DF PV PN SV MB
HS-SPME-GC–MS 47 Diethyl succinate 0.00 0.00 0.00 0.00 0.00 0.05
48 Ethyl 2-methylbutyrate 6.94 6.94 8.40 2.54 12.42 0.03
49 Ethyl isovalerate 1.06 1.72 0.47 2.65 1.50 0.02
50 Ethyl nonanoate 0.00 0.00 0.00 0.00 0.00 0.02
51 Methyl salicylate 0.00 0.00 0.00 0.00 0.00 0.01
52 Isopentyl hexanoate 0.00 0.00 0.00 0.00 0.00 0.03
53 Geranyl acetate – – – – – 0.01
54 Terpinen-4-ol 45.08 26.20 53.41 2.40 2.01 1.81
55 Nerol 0.07 0.03 0.09 0.01 0.03 0.05
56 (E, E)-Farnesol 17.38 14.40 7.14 6.94 8.98 0.06
57 Geraniol 0.43 0.34 0.59 0.28 0.36 0.02
58 d-Limonene 0.23 0.35 0.21 0.40 0.35 0.01
59 α-Terpineol 0.00 0.00 0.00 n.d. 0.00 0.02
60 Linalool 0.08 0.05 0.07 0.08 0.08 0.02
61 Citronellol 0.00 0.00 0.00 0.00 0.00 0.01
62 Rose oxide 0.59 0.20 0.78 0.37 0.49 0.01
63 β-Lonone 8.92 8.76 7.19 7.20 10.05 0.01
64 3-Hydroxy-2-butanone 6.50 6.61 6.88 0.50 3.84 0.65
65 Phenol – – – – – 0.04
66 4-Ethylphenol 0.00 0.00 0.00 0.00 0.00 0.01
67 Furfural 0.00 0.00 0.00 0.00 0.00 0.05
68 Decanal 0.00 0.00 0.00 0.00 0.00 0.01
69 Octanal – – – – – 0.02
Identification methods Number Compounds OAV/ROAV
 VIP pred
DF PV PN SV MB
HS-SPME-GC–MS 70 Benzaldehyde 0.00 0.00 0.00 0.00 0.00 0.01
HS-GC-IMS 71 Acetic acid 0.00 0.00 0.00 0.00 0.00 1.92
72 2,3-Dimethylpyrazine 0.00 0.00 0.00 0.00 0.00 0.75
73 2,4,6-Trimethylpyridine – – – – – 0.99
74 Octanal – – – – – 1.02
75 Ethyl hexanoate 0.02 0.02 0.02 0.02 0.03 0.49
76 3-Methyl-3-butenol – – – – – 0.53
77 Hexyl acetate 0.00 0.00 0.00 0.00 0.00 0.44
78 Cyclopentanone – – – – – 0.69
79 4-Methyl-2-pentanol – – – – – 1.73
80 β-Myrcene – – – – – 0.21
81 Isoamyl acetate 0.08 0.09 0.17 0.11 0.10 0.93
82 2-Methylbutyl acetate 1.00 0.96 0.66 0.67 1.00 2.42
83 3-Pentanol – – – – – 0.54
84 4-Methyl-3-penten-2-one – – – – – 0.15
85 Hexanal 0.00 0.00 0.00 0.00 0.00 0.42
86 2-Hexanone – – – – – 0.74
87 Linalool oxide – – – – – 0.34
88 (E)-2-Octenal 0.77 1.00 1.00 1.10 1.14 0.74
89 2-Methyl-1-propanol 0.00 0.00 0.00 0.00 0.00 0.58
90 Ethyl isovalerate 0.14 0.40 0.09 0.55 0.41 0.99
91 Ethyl 2-methylbutyrate 0.49 0.68 0.49 0.90 0.74 0.67
92 2-Butanol 0.00 0.00 0.00 0.00 0.00 0.56
Identification methods Number Compounds OAV/ROAV
 VIP pred
DF PV PN SV MB
HS-GC-IMS 93 Ethyl butyrate 0.62 0.58 0.62 0.48 0.69 1.06
94 Methyl 2-methylbutanoate – – – – – 0.91
95 4-Methyl 2-pentanone – – – – – 0.44
96 Isobutyl acetate 0.00 0.01 0.00 0.00 0.01 1.66
97 2,3-Butanedione 0.01 0.00 0.01 0.01 0.00 0.49
98 2-Pentanone – – – – – 0.66
99 Propyl acetate 0.00 0.00 0.00 0.00 0.00 0.88
100 Ethyl isobutyrate 0.09 0.24 0.09 0.18 0.24 0.92
101 Ethyl propanoate 0.00 0.00 0.00 0.00 0.00 1.01
102 2-Propanol – – – – – 1.08
103 Ethyl acetate 0.01 0.00 0.01 0.00 0.01 1.17
104 Methyl acetate – – – – – 1.28
105 Propanal – – – – – 0.57
106 Pentanal – – – – – 1.50
Open in a new tab

--: not be calculated, odor thresholds for relevant compounds were not found. n.d.: not detected.

The odor thresholds and aroma descriptions of compounds are shown in Appendix Table S5.

3.5. Screening of potential characteristic compounds based on ML algorithms

Four ML algorithms (NB, RF, KNN, DT) were used to construct models, with a training-to-test set ratio of 7:3. The performance of each model on the test set is presented in Table S6 (HS-SPME-GC–MS) and Table S7 (HS-GC-IMS). The ROC curves are shown in Fig. S2 and Fig. S3. The AUC measures the model's ability to distinguish between positive and negative classes, with an AUC of 1 indicating a perfect model. The F1 score is the harmonic mean of precision and recall, ranging from 0 to 1, where 1 is the best possible score, representing perfect precision and recall, and 0 is the worst (Zhou et al., 2025). Based on the HS-SPME-GC–MS data, NB and RF models demonstrated excellent performance (AUC = 1.000, F1 = 1.000). Based on the HS-GC-IMS data, RF model showed excellent performance (AUC = 1.000, F1 = 1.000). Therefore, the RF model was used to screen compounds. The top 20 VOCs in terms of importance are shown in Fig. 5.

Fig. 5.

Fig. 5

Open in a new tab

Importance Map of VOCs Screened by RF model based on HS-SPME-GC–MS (A) and HS-GC-IMS (B) Data. The figure presents the top 20 VOCs in five wines.

According to RF analysis results, the top 5 compounds were selected as potential characteristic compounds. Combined with OAV > 1 or ROAV ≥1 to screen out the key differential VOCs, we found that ethyl hexanoate, ethyl caprate and phenethyl acetate might be the main factors for differences between DF and others, 3-methyl-butanoic acid and isoamyl acetate were the key substances that distinguish PV from others, 2-methoxy-3-isobutyl pyrazine is the key differential VOC for PN, ethyl hexanoate is the key differential VOC for SV, and terpinen-4-ol and β-lonone may be the main factors that cause MB to be different from others. Other compounds did not make significant contributions to overall aroma of wines due to their relatively high odor thresholds. Therefore, we consider that 8 compounds, namely ethyl hexanoate, ethyl caprate, phenethyl acetate, 3-methyl-butanoic acid, isoamyl acetate, 2-methoxy-3-isobutyl pyrazine, terpinen-4-ol and β-lonone are the key VOCs that distinguish the flavor differences among five wines.

OPLS-DA relies on the principle of multivariate statistical analysis and achieves feature screening by maximizing the differences between groups. In contrast, RF constructs multiple decision trees and screens features based on their contributions to the classification results (Vu et al., 2019). There are fundamental differences in their principles and data processing methods, which result in fact that the characteristic compounds screened by them are not completely consistent. In view of this, when researching the differentiation and discrimination of different wines, we can pay attention to and analyze these characteristic differences based on the screening results of these two methods.

3.6. Correlation analysis of specific response of E-nose with key VOCs

Based on OPLS-DA and RF models combined with OAV > 1 or ROAV ≥1, a total of 15 key differential VOCs were identified in wines of different varieties. Fig. 6A shows the distribution heatmap of these 15 key differential VOCs. We found that phenylethyl alcohol, ethyl acetate and 3-methyl-1-butanol were the key differential compounds in SV; (Z)-2-hexen-1-ol, (E)-2-hexen-1-ol, ethyl hexanoate, ethyl caprate and phenethyl acetate were the key differential compounds in DF; terpinen-4-ol and 2-methoxy-3-isobutyl pyrazine were the key differential compounds in PN; 2-methyl-1-propanol and β-lonone were the key differential compounds in MB. Additionally, 2-methylbutyl acetate, 3-methyl-butanoic acid and isoamyl acetate were identified as key differential compounds for distinguishing PV. Meanwhile, mathematical correlations between E-nose sensor responses (aroma attributes) and key differential VOCs were established by Spearman analysis. As shown in Fig. 6B, W1S was significantly positively correlated (p < 0.05) with isoamyl acetate, W1W was significantly positively correlated (p < 0.05) with 3-methyl-butanoic acid, W2S was significantly positively correlated (p < 0.05) with isoamyl acetate and extremely significantly positively correlated (p < 0.01) with 2-methylbutyl acetate. W2W was extremely significantly positively correlated (p < 0.01) with 3-methyl-butanoic acid and significantly positively correlated (p < 0.05) with 3-methyl-1-butanol. W5S was extremely significantly positively correlated (p < 0.01) with isoamyl acetate and significantly positively correlated (p < 0.05) with phenethyl acetate and 2-methylbutyl acetate. These correlations indicate that W1S, W2S and W5S had strong response capabilities to esters, while W1W and W2W were more sensitive to organic acids and alcohols. Additionally, most of these compounds were found to be highly volatile and low-boiling substances (isoamyl acetate, 2-methylbutyl acetate, 3-methyl-1-butanol), which is consistent with findings of Xu et al. (2023). As a simulated analytical tool for sensory evaluation, E-nose can effectively distinguish the aroma characteristics of samples, a conclusion supported by numerous similar studies (Chen et al., 2020; Gerhardt et al., 2019; Xiao et al., 2022). In summary, different samples had their own key differential VOCs, and specific correlations existed between E-nose sensors and these compounds. Through rapid E-nose analysis, the levels of 2-methylbutyl acetate (W2S), 3-methylbutanoic acid (W2W), and isoamyl acetate (W5S) could be better predicted (p < 0.01), thus quickly distinguishing PV from the other wines.

Fig. 6.

Fig. 6

Open in a new tab

Heat maps of distribution of key differential VOCs levels in wines of five varieties (A) and heat maps of correlation between E-nose sensors and key VOCs (B). The numbers in Fig. 6A are results of normalization of original data within the range of 0–1. Spearman correlation analysis: asterisks (*) denote significance of correlations，* represents p < 0.05 and ** represents p < 0.01.

4. Conclusions

In this study, wines made from five grape varieties harvested from Chateau Greatwall Terroir (Ningxia, China) were comprehensively analyzed for VOCs by E-nose, HS-SPME-GC–MS and HS-GC-IMS. The results showed that there were significant differences (p < 0.05) in aroma attributes and VOC compositions among five wines. A total of 94 VOCs were identified by HS-SPME-GC–MS and HS-GC-IMS, with 70 and 36 VOCs identified by each method respectively, among which 12 compounds were detected by both methods. HS-SPME-GC–MS was proved to be more sensitive in identifying alcohols (their concentration accounting for 52.56%–68.75 % of all 70 VOCs), while HS-GC-IMS was more sensitive in esters (their concentration accounting for 35.58 %–42.05 % of all 36 VOCs). Cluster analysis classified the five wines into three categories: SV, PV & MB and PN & DF. Based on chemometrics (OPLS-DA, OAV/ROAV, VIP) and ML (RF) models analysis, 15 VOCs were identified to make significant contributions to differentiating the aroma characteristics of five wines, including phenylethyl alcohol, (Z)-2-hexen-1-ol, (E)-2-hexen-1-ol, ethyl acetate, terpinen-4-ol, 3-methyl-1-butanol, 2-methyl-1-propanol, 2-methylbutyl acetate, ethyl hexanoate, ethyl caprate, phenethyl acetate, 3-methyl-butanoic acid, isoamyl acetate, 2-methoxy-3-isobutyl pyrazine and β-lonone. The sensors with higher response values from E-nose were W1S, W1W, W2S, W2W and W5S. Additionally, W2S, W2W and W5S showed extremely significant positive correlations (p < 0.01) with the levels of 2-methylbutyl acetate, 3-methyl-butanoic acid and isoamyl acetate, respectively, enabling rapid differentiation of PV from others.

This study reveals the characteristics and differences of different wines in terms of aroma composition by identifying the key differential compounds, which can provide a chemical basis for sample differentiation and quality assessment. Future research will integrate multi-source data such as the inorganic components, organic components and isotope information of wines, as well as sensory evaluation and origin environmental data. The ML model will be further optimized to improve the accuracy of screening characteristic compounds and the precision of wine discrimination, providing a more scientific theoretical basis for optimization of wine brewing processes, quality control and development of new products.

CRediT authorship contribution statement

Rui Xie: Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Jiawen Liu: Visualization, Investigation, Data curation. Yutao Li: Visualization, Software, Data curation. Yong Chen: Investigation, Formal analysis. Tian Shen: Investigation. Meilong Xu: Supervision. Yanlun Ju: Supervision, Funding acquisition, Conceptualization. Yulin Fang: Project administration, Methodology, Funding acquisition. Zhenwen Zhang: Writing – review & editing, Supervision, Resources.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

We are grateful for the support from the Chateau Greatwall Terroir in Ningxia, China and Ms. Jing Zhang, Ms. Jing Zhao and Ms. WenJing Cao (Horticulture Science Research Center, Northwest A&F University, Yangling, China) for providing professional technical assistance with HS-GC-IMS analysis. This work was supported by the Ningxia Hui Autonomous Region Key Research and Development Program (2024BBF01002-04) and China Agriculture Research System for Grape Industry (CARS-29-zp-6).

Footnotes

Appendix A

Supplementary data to this article can be found online at https://doi.org/10.1016/j.fochx.2025.103082.

Contributor Information

Rui Xie, Email: xierui@nwafu.edu.cn.

Yanlun Ju, Email: juyanlun2016@nwsuaf.edu.cn.

Yulin Fang, Email: fangyulin@nwsuaf.edu.cn.

Zhenwen Zhang, Email: zhangzhw60@nwsuaf.edu.cn.

Appendix A. Supplementary data

Supplementary material
mmc1.docx (1.3MB, docx)

Data availability

Data will be made available on request.

References

  1. Álvarez-Fernández M.A., Carafa I., Vrhovsek U., Arapitsas P. Modulating wine aromatic amino acid catabolites by using Torulaspora delbrueckii in sequentially inoculated fermentations or Saccharomyces cerevisiae alone. Microorganisms. 2020;8(9):1349. doi: 10.3390/microorganisms8091349. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Baldwin E.A., Bai J., Plotto A., Dea S. Electronic noses and tongues: Applications for the food and pharmaceutical industries. Sensors. 2011;11(5):4744–4766. doi: 10.3390/s110504744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Cao W.Y., Shu N., Wen J.L., Yang Y.M., Jin Y.N., Lu W.P. Characterization of the key aroma volatile compounds in nine different grape varieties wine by headspace gas chromatography–ion mobility spectrometry (HS-GC-IMS), odor activity values (OAV) and sensory analysis. Foods. 2022;11(18):2767. doi: 10.3390/foods11182767. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Chen X.A., Chen H.Q., Xiao J., Liu J.Y., Tang N., Zhou A.M. Variations of volatile flavor compounds in finger citron (Citrus medica L. var. sarcodactylis) pickling process revealed by E-nose, HS-SPME-GC-MS and HS-GC-IMS. Food Research International. 2020;138 doi: 10.1016/j.foodres.2020.109717. [DOI] [PubMed] [Google Scholar]
  5. Chen Y., Li P., Liao L.Y., Qin Y.Y., Jiang L.W., Liu Y. Characteristic fingerprints and volatile flavor compound variations in Liuyang Douchi during fermentation via HS-GC-IMS and HS-SPME-GC-MS. Food Chemistry. 2021;361 doi: 10.1016/j.foodchem.2021.130055. [DOI] [PubMed] [Google Scholar]
  6. Escudero A., Campo E., Fariña L., Cacho J., Ferreira V. Analytical characterization of the aroma of five premium red wines. Insights into the role of odor families and the concept of fruitiness of wines. Journal of Agricultural and Food Chemistry. 2007;55(11):4501–4510. doi: 10.1021/jf0636418. [DOI] [PubMed] [Google Scholar]
  7. Feng T., Sun J.Q., Song S.Q., Wang H.T., Yao L.Y., Sun M.…Chen D. Geographical differentiation of Molixiang table grapes grown in China based on volatile compounds analysis by HS-GC-IMS coupled with PCA and sensory evaluation of the grapes. Food Chemistry: X. 2022;15 doi: 10.1016/j.fochx.2022.100423. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Fia G., Gori C., Bucalossi G., Borghini F., Zanoni B.A. Naturally occurring antioxidant complex from unripe grapes: The case of Sangiovese (v. Vitis vinifera) Antioxidants. 2018;7(2):27. doi: 10.3390/antiox7020027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Gerhardt N., Schwolow S., Rohn S., Pérez-Cacho P.R., Galán-Soldevilla H., Arce L., Weller P. Quality assessment of olive oils based on temperature-ramped HS-GC-IMS and sensory evaluation: Comparison of different processing approaches by LDA, KNN, and SVM. Food Chemistry. 2019;278:720–728. doi: 10.1016/j.foodchem.2018.11.095. [DOI] [PubMed] [Google Scholar]
  10. Jiang B., Zhang Z.W. A preliminary study of aroma composition and impact odorants of cabernet franc wines under different terrain conditions of the loess plateau region (China) Molecules. 2018;23(5):1096. doi: 10.3390/molecules23051096. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Jin W.G., Zhao P., Liu J.X., Geng J.Z., Chen X.H., Pei J.J., Jiang P.F. Volatile flavor components analysis of giant salamander (Andrias davidiauns) meat during roasting process based on gas chromatography-ion mobility spectroscopy and chemometrics. Food and Fermentation Industries. 2021;47(21):231–239. doi: 10.13995/j.cnki.11-1802/ts.028024. [DOI] [Google Scholar]
  12. Kong C.L., Li A.H., Su J., Wang X.C., Chen C.Q., Tao Y.S. Flavor modification of dry red wine from Chinese spine grape by mixed fermentation with Pichia fermentans and S. Cerevisiae. LWT - Food Science and Technology. 2019;109:83–92. doi: 10.1016/j.lwt.2019.03.101. [DOI] [Google Scholar]
  13. Li Q.Q., Zhang C.Y., Liu W., Li B., Chen S.F., Wang H.W.…Li J.X. Characterization and exploration of dynamic variation of volatile compounds in vine tea during processing by GC-IMS and HS-SPME/GC–MS combined with machine learning algorithm. Food Chemistry. 2024;460 doi: 10.1016/j.foodchem.2024.140580. [DOI] [PubMed] [Google Scholar]
  14. Li W.P., Yao H., Chen K.Q., Ju Y.L., Min Z., Sun X.Y.…Fang Y.L. Effect of foliar application of fulvic acid antitranspirant on sugar accumulation, phenolic profiles and aroma qualities of cabernet sauvignon and Riesling grapes and wines. Food Chemistry. 2021;351 doi: 10.1016/j.foodchem.2021.129308. [DOI] [PubMed] [Google Scholar]
  15. Li Y., Zhang Y., Wang Y.Y., Li X., Zhou L., Yang J., Guo L.P. Metabolites and chemometric study of Perilla (Perilla frutescens) from different varieties and geographical origins. Journal of Food Science. 2022;87(12):5240–5251. doi: 10.1111/1750-3841.16376. [DOI] [PubMed] [Google Scholar]
  16. Liu R., Liu Y., Zhu Y., Kortesniemi M., Zhu B., Li H. Aromatic characteristics of passion fruit wines measured by E-nose, GC-quadrupole MS, GC-orbitrap-MS and sensory evaluation. Foods. 2022;11(23):3789. doi: 10.3390/foods11233789. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Noguerol-Pato R., González-Álvarez M., González-Barreiro C., Cancho-Grande B., Simal-Gándara J. Aroma profile of Garnacha Tintorera-based sweet wines by chromatographic and sensorial analyses. Food Chemistry. 2012;134(4):2313–2325. doi: 10.1016/j.foodchem.2012.03.105. [DOI] [PubMed] [Google Scholar]
  18. Peinado R.A., Moreno J., Bueno J.E., Moreno J.A., Mauricio J.C. Comparative study of aromatic compounds in two young white wines subjected to pre-fermentative cryomaceration. Food Chemistry. 2004;84(4):585–590. doi: 10.1016/S0308-8146(03)00282-6. [DOI] [Google Scholar]
  19. Rodríguez-Maecker R., Vyhmeister E., Meisen S., Rosales Martinez A., Kuklya A., Telgheder U. Identification of terpenes and essential oils by means of static headspace gas chromatography-ion mobility spectrometry. Analytical and Bioanalytical Chemistry. 2017;409:6595–6603. doi: 10.1007/s00216-017-0613-2. [DOI] [PubMed] [Google Scholar]
  20. Shinohara T. Gas chromatographic analysis of volatile fatty acids in wines. Agricultural and Biological Chemistry. 1985;49(7):2211–2212. doi: 10.1080/00021369.1985.10867054. [DOI] [Google Scholar]
  21. Stranska M., Uttl L., Bechynska K., Hurkova K., Behner A., Hajslova J. Metabolomic fingerprinting as a tool for authentication of grapevine (Vitis vinifera L.) biomass used in food production. Food Chemistry. 2021;361 doi: 10.1016/j.foodchem.2021.130166. [DOI] [PubMed] [Google Scholar]
  22. Sun X.Y., Ma T.T., Yu J., Huang W.D., Fang Y.L., Zhan J.C. Investigation of the copper contents in vineyard soil, grape must and wine and the relationship among them in the Huaizhuo Basin region, China: A preliminary study. Food Chemistry. 2018;241:40–50. doi: 10.1016/j.foodchem.2017.08.074. [DOI] [PubMed] [Google Scholar]
  23. Vecchio R., Decordi G., Grésillon L., Gugenberger C., Mahéo M., Jourjon F. European consumers' perception of moderate wine consumption on health. Wine Economics and Policy. 2017;6(1):14–22. doi: 10.1016/j.wep.2017.04.001. [DOI] [Google Scholar]
  24. Vu T., Siemek P., Bhinderwala F., Xu Y.H., Powers R. Evaluation of multivariate classification models for analyzing NMR metabolomics data. Journal of Proteome Research. 2019;18(9):3282–3294. doi: 10.1021/acs.jproteome.9b00227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Wang L., Chen Z., Han B., Wu W.X., Zhao Q.L., Wei C.Q., Liu W.Y. Comprehensive analysis of volatile compounds in cold-pressed safflower seed oil from Xinjiang, China. Food Science & Nutrition. 2020;8(2):903–914. doi: 10.1002/fsn3.1369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Wang Y.Y., Ma Y.H., Duan J.L., Wang B., Ma T.Z., Jiang Y.M., Zhang B. Discrimination and characterization of the volatile organic compounds in red and black raspberry wines fermented with different commercial Saccharomyces cerevisiae: An integrated analysis using E-nose, GC-MS, GC-IMS, and multivariate statistical models. Food Chemistry. 2025 doi: 10.1016/j.foodchem.2025.143678. [DOI] [PubMed] [Google Scholar]
  27. Wei G.Z., Dan M.L., Zhao G.H., Wang D.M. Recent advances in chromatography-mass spectrometry and electronic nose technology in food flavor analysis and detection. Food Chemistry. 2023;405 doi: 10.1016/j.foodchem.2022.134814. [DOI] [PubMed] [Google Scholar]
  28. Xi B.N., Zhang J.J., Xu X., Li C., Shu Y., Zhang Y.…Shen Y.H. Characterization and metabolism pathway of volatile compounds in walnut oil obtained from various ripening stages via HS-GC-IMS and HS-SPME-GC–MS. Food Chemistry. 2024;435 doi: 10.1016/j.foodchem.2023.137547. [DOI] [PubMed] [Google Scholar]
  29. Xiao Y., Huang Y.X., Chen Y.L., Zhu M.Z., He C., Li Z.J.…Liu Z.H. Characteristic fingerprints and change of volatile organic compounds of dark teas during solid-state fermentation with Eurotium cristatum by using HS-GC-IMS, HS-SPME-GC-MS, E-nose and sensory evaluation. LWT - Food Science and Technology. 2022;169 doi: 10.1016/j.lwt.2022.113925. [DOI] [Google Scholar]
  30. Xu C.C., Li Q.L., Wang N., Liu D.K., Guo C.X. Identifying and discriminating aroma attribute and bioactive components of five commercial essential oils of celery (Apium graveolens L.) seeds using E-nose, HS-GC-IMS, and GC-MS. LWT - Food Science and Technology. 2023;184 doi: 10.1016/j.lwt.2023.115094. [DOI] [Google Scholar]
  31. Yan J., Wang H., Wang Y., Xu S., Wan Y.H., He L.F.…Zhu W.M. Integrated metabolome and transcriptome analysis reveals candidate genes involved in metabolism of terpenoids and phthalides in celery seeds. Industrial Crops and Products. 2021;172 doi: 10.1016/j.indcrop.2021.114011. [DOI] [Google Scholar]
  32. Yang W.X., You Y.Z., Ling M.Q., Ye D.Q., Shi Y., Duan C.Q., Lan Y.B. Identification of the key odor-active compounds responsible for varietal smoky aroma in wines made from the east Asian species. Food Research International. 2023;171 doi: 10.1016/j.foodres.2023.113052. [DOI] [PubMed] [Google Scholar]
  33. Yin H.N., Wang L., Su H.M., Liang Y.Y., Ji P., Wang X.F., Xi Z.M. Effects of ultraviolet and infrared radiation absence or presence on the aroma volatile compounds in winegrape during veraison. Food Research International. 2023;167 doi: 10.1016/j.foodres.2023.112662. [DOI] [PubMed] [Google Scholar]
  34. Yun J., Cui C.J., Zhang S.H., Zhu J.J., Peng C.Y., Cai H.M.…Hou R.Y. Use of headspace GC/MS combined with chemometric analysis to identify the geographic origins of black tea. Food Chemistry. 2021;360 doi: 10.1016/j.foodchem.2021.130033. [DOI] [PubMed] [Google Scholar]
  35. Zeng X.Q., Cao R., Xi Y., Li X.J., Yu M.H., Zhao J.L.…Li J. Food flavor analysis 4.0: A cross-domain application of machine learning. Trends in Food Science & Technology. 2023;138:116–125. doi: 10.1016/j.tifs.2023.06.011. [DOI] [Google Scholar]
  36. Zhang L., Tao Y.S., Wen Y., Wang H. Aroma evaluation of young Chinese merlot wines with denomination of origin. South African Journal of Enology and Viticulture. 2013;34(1):46–53. doi: 10.1515/ijfe-2012-0202. [DOI] [Google Scholar]
  37. Zhao T., Wu J.Y., Meng J.F., Shi P.B., Fang Y.L., Zhang Z.W., Sun X.Y. Harvesting at the right time: Maturity and its effects on the aromatic characteristics of cabernet sauvignon wine. Molecules. 2019;24(15):2777. doi: 10.3390/molecules24152777. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Zhou C.L., Yu Y.S., Ai J.Y., Song C., Cui Z.Y., Zhou Q.L.…Liu Y. Fruit wines classification enabled by combing machine learning with comprehensive volatiles profiles of GC-TOF/MS and GC-IMS. Food Research International. 2025 doi: 10.1016/j.foodres.2025.115890. [DOI] [PubMed] [Google Scholar]
  39. Zhou Y.X., Wang D.D., Duan H., Zhou S.Q., Guo G.H., Yan W.J. Detection and analysis of volatile flavor compounds in different varieties and origins of goji berries using HS-GC-IMS. LWT - Food Science and Technology. 2023;187 doi: 10.1016/j.lwt.2023.115322. [DOI] [Google Scholar]
  40. Zhu J.C., Chen F., Wang L.Y., Niu Y.W., Xiao B.Z. Evaluation of the synergism among volatile compounds in oolong tea infusion by odour threshold with sensory analysis and E-nose. Food Chemistry. 2017;221:1484–1490. doi: 10.1016/j.foodchem.2016.11.002. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary material
mmc1.docx (1.3MB, docx)

Data Availability Statement

Data will be made available on request.

Articles from Food Chemistry: X are provided here courtesy of Elsevier

RESOURCES