Browsing by Person "Zhang, Yanyan"

Now showing 1 - 12 of 12

Characterization of aroma and sensorial variation of basil and ginger during short-time drying process
(2024) Liang Jiaqi; Zhang, Yanyan
Industries are increasingly interested in aromatic herbs and spices, such as basil and ginger, which play a crucial role in daily life as flavoring agents in foods, beverages, and pharmaceuticals. Basil, an annual herb from the mint family (Lamiaceae), is renowned for its wonderful “royal” fragrance and numerous health benefits. Ginger, one of the most widely consumed dietary spices worldwide, has seen a surge in interest due to its significant healthpromoting properties. In 2023, global market sales for basil leaves and ginger reached 50 million euros and 4.2 billion euros, respectively. As may other herbs and spices, basil and ginger are typically used in dried form to extend shelf life and facilitate transportation. However, the drying process significantly alter the aroma from fresh samples, reduce the quality of the final products and generate numerous by-products. This study aims to identify the off-odor compound(s) produced during basil drying process using molecular sensory techniques and to develop effective strategies to inhibit their formation. Additionally, ginger peel and ginger fiber by-products generated during the industrial and domestic use, as well as the spray drying process, were investigated and reintroduced into the procedure to preserve the aroma compounds of ginger and promote environmental sustainability. In the first part of the study, a method using stir bar sorptive extraction combined with gas chromatography-mass spectrometry-olfactory (SBSE-GC-MS-O) was established to analyse the aroma changes of basil and ginger during short-term spray drying. This approach was designed to maximize desorption efficiency for various odorants with diverse physicochemical properties under different desorption conditions through four types of mathematical modelling. Among these models, the Random Forest model demonstrated the highest performance and minimal errors, with an R value of 0.910 after validation using a dataset of six new compounds. In addition, the model determined that cryo-focusing temperature was the most important factor, followed by molecular weight, log P, boiling point, desorption temperature, desorption time, and helium flow. This algorithm can be further utilized to predict the optimal parameters for maximizing desorption efficiency in aroma analysis of basil and ginger by SBSE-GC-MS-O. Subsequently, to investigate the formation of the hay-like off-odor that lowers consumer acceptability in dried basil products, the responsible compound was identified in thawed, airdried, and spray-dried basil samples using a trained human panel (n = 10) and gas chromatography–mass spectrometry–olfactometry. 3-Methylnonane-2,4-dione (3-MND) was found to be the odorant contributing to the hay-like off-odor in all basil samples. In order to reduce this odorant, the effects of light, oxygen, and temperature on the formation of 3-MND during the processes of thawing, spray drying, and air drying were studied based on a potential pathway involving 3-MND precursors. The results revealed that controlling light exposure, employing nitrogen protective environment, and maintaining low temperatures are critical processing parameters for minimizing the generation of the hay-like compound 3-MND, thereby meeting consumer demands for high-quality dried basil products. Regarding the processing of ginger, the drying process generates a significant amount of ginger peel as an industrial by-product. To facilitate environmental sustainability and establish a reference for its potential applications, the effect of ginger peel on aroma, sensory profiles, and nutrition-related physicochemical properties was investigated. The total concentration of aroma compounds in unpeeled ginger was 1.3 times higher than that in peeled ginger, according to SBSE-GC-MS-O analysis. Sensory evaluation data indicated that unpeeled ginger had significantly enhanced citrus-like and fresh impressions compared to peeled ginger, which is associated with the higher odor activity values of odorants such as β-myrcene (pungent, citruslike), geranial (citrus-like), citronellal (citrus-like, sourish), and linalool (floral, fresh). Additionally, the total polyphenol content in unpeeled ginger was 7.96 mg/100 g higher than that in peeled ginger. Unpeeled ginger not only demonstrates more intense aromatic and sensory qualities but also offers superior nutritional and environmental benefits, making it a promising option for future global spice use and reducing ginger side streams. In the spray drying process of ginger, ginger fibre is also typically regarded as industrial byproduct, as ginger juice is often used as a feed material. To achieve sustainable development and produce a clean-labeled product, a split-stream spray-drying process was developed to address sugar-rich feed solutions by reintroducing ginger fiber in their natural composition as a carrier material, rather than treating them as a by-product. The characterizations of both feed materials were then compared. The method was optimized for aroma retention by adjusting the inlet and outlet temperatures to 220 ℃ and 80 ℃, respectively. Aroma decoding results using SBSE-GC-MS-O demonstrated that reintegrating ginger fiber significantly increased the concentration of eight key odorants, including hexanal, linalool, neral, borneol, geranial, citronellol, nonanoic acid, and α-bisabolol, in comparison to the concentration observed in ginger juice. This presents a promising solution for maximizing the utilization of ginger in spray drying, enhancing the aroma profiles of feed material, and addressing sustainability considerations in the food industry. Overall, the presented work comprehensively explores the challenges and potential solutions associated with the drying process of basil and ginger, highlighting the impact of various processing conditions on aroma and sensory properties. By identifying key off-odor compounds and optimizing techniques to reduce their formation, as well as investigating aroma profiles and sustainable approaches for utilizing by-products such as ginger peel and ginger fiber, this research offers valuable insights into improving both the aromatic quality and environmental sustainability of dried herb and spice products.
Characterization of the aroma profile of food smoke at controllable pyrolysis temperatures
(2023) Rigling, Marina; Höckmeier, Laura; Leible, Malte; Herrmann, Kurt; Gibis, Monika; Weiss, Jochen; Zhang, Yanyan
Smoking is used to give food its typical aroma and to obtain the desired techno-functional properties of the product. To gain a deeper knowledge of the whole process of food smoking, a controllable smoking process was developed, and the influence of wood pyrolysis temperature (150–900 °C) on the volatile compounds in the smoking chamber atmosphere was investigated. The aroma profile of smoke was decoded by headspace solid-phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS). Subsequently, the correlations in the most important substance classes, as well as in individual target components, were investigated by the Pearson test. Phenols and lactones showed an increase over the entire applied temperature range (rT = 0.94 and rT = 0.90), whereas furans and carbonyls showed no strict temperature dependence (rT < 0.6). Investigations on single aroma compounds showed that not all compounds of one substance class showed the same behavior, e.g., guaiacol showed no significant increase over the applied pyrolysis temperature, whereas syringol and hydoxyacetone showed a plateau after 450 °C, and phenol and cyclotene increased linear over the applied temperature range. These findings will help to better understand the production of aroma-active compounds during smoke generation in order to meet consumers preferences.
Characterization of the aroma properties in fragrant rapeseed oil and aroma variation during critical roasting phase
(2023) Zhang, Youfeng; Zhang, Yanyan
Rapeseed oil is one of the third most-produced vegetable oil in the world, which is appreciated for its characteristic flavor and high nutritional value. Fragrant rapeseed oil (FRO) produced by a typical roasting process is popular for its characteristic aroma, which has an annual consumption exceeding 1.5 million tons. However, the changes in aroma blueprint of FRO during the typical roasting processing are still unclear, which challenges rapeseed oil quality and consumer acceptance. Accordingly, the aim of this work was to investigate the aroma characteristics and their precursors pyrolysis behavior of FRO to provide a basis and guidance for the control of FRO aroma quality during production processing. First, a systematic review on summarizing, comparing, and critiquing the literature regarding the flavor of rapeseed oil, especially about employed analysis techniques (i.e., extraction, qualitative, quantitative, sensorial, and chemometric methods), identified representative/off-flavor compounds, and effects of different treatments during the processes (dehulling, roasting, microwave, flavoring with herbs, refining, oil heating, and storage) was performed. One hundred and thirty-seven odorants found in rapeseed oil from literature are listed, including aldehydes, ketones, acids, esters, alcohols, phenols, pyrazines, furans, pyrrolines, indoles, pyridines, thiazoles, thiophenes, further S-containing compounds, nitriles, and alkenes, and possible formation pathways of some key aroma-active compounds are also proposed. Nevertheless, some of these compounds require further validation (e.g., nitriles) due to lack of recombination experiments in the previous work. To wrap up, advanced flavor analysis techniques should be evolved toward time-saving, portability, real-time monitoring, and visualization, which aims to obtain a “complete” flavor profile of rapeseed oil. Aparting from that, studies to elucidate the influence of key roasting processing on the formation of aroma-active compounds are needed to deepen understanding of factors resulting in flavor variations of rapeseed oil. Following, a systematic comparison among five flavor trapping techniques including solid-phase microextraction (SPME), SPME-Arrow, headspace stir bar sorptive extraction (HSSE), direct thermal desorption (DTD), and solvent-assisted flavor evaporation (SAFE) for hot-pressed rapeseed oil was conducted. Besides, methodological validation of these five approaches for 31 aroma standards found in rapeseed oil was conducted to compare their stability, reliability, and robustness. For the qualification of the odorants in hot-pressed rapeseed oil, SAFE gave the best performance, mainly due to the high sample volumes, but it performed worse than other methods regarding linearity, recovery, and repeatability. SPME-Arrow gave good performances in not only odorant extraction but also quantification, which is considered most suitable for quantifying odorants in hot-pressed rapeseed oil. Taking cost/performance ratio into account, SPME is still an efficient flavor extraction method. Multi-method combination of flavor capturing techniques might also be an option of aroma analysis for oil matrix. Afterwards, by application of the Sensomics approach the key odorants in representative commercial FRO samples were decoded. On the basis of the aroma blueprint, changes of overall aroma profiles of oils and their key odorants were studied and compared in different roasting conditions. To better simulate industrial conditions, high temperatures (150-200 ºC) were used in our roasting study, which was rarely studied before. Identification and quantitation of the key odorants in FRO were well performed by means of the Sensomics concept. Glucosinolate degradation products were a special kind of key odorants existing in rapeseed oil. Most of the odorants showed first rising and then decline trends as the roasting process progressed. Aroma profile results showed that high-temperature-short time and low-temperature-long time conditions could have similar effects on the aroma profiles of roasted rapeseed oils, which could provide a reference for the time cost savings in industrial production. To gain the fundamental knowledge of the aroma formation in FRO, the thermal degradation behavior of progoitrin (the main glucosinolate of rapeseed) and the corresponding generated volatile products were investigated in liquid (phosphate buffer at a pH value of 5.0, 7.0, or 9.0) and solid phase systems (sea sand and rapeseed powder). The highest thermal degradation rate of progoitrin at high temperatures (150-200 ºC) was observed at a pH value of 9.0, followed by sea sand and then rapeseed powder. It could be inferred that bimolecular nucleophilic substitution reaction (SN2) was mainly taken place under basic conditions. The highest degradation rate under basic conditions might result from the high nucleophilicity of present hydroxide ions. Under the applied conditions in this study, 2,4-pentadienenitrile was the major nitrile formed from progoitrin during thermal degradation at high temperature compared to l-cyano-2-hydroxy-3-butene, which might be less stable. The possible formation pathways of major S-containing (thiophenes) and N-containing (nitriles) volatile (flavor) compounds were proposed. Hydrogen sulfide, as a degradation product of glucosinolates, could act as a sulfur source to react further with glucose to generate thiophenes. Overall, the present work comprehensively documented the effects of thermal conditions and matrices on the aroma characteristics, aroma profiles, and key odorants of hot-pressed rapeseed oil, which could provide data and theoretical basis for the flavor control of FRO under thermal treatment at actual production temperatures (150-200 °C).
Coffee leaf tea from El Salvador: on-site production considering influences of processing on chemical composition
(2022) Steger, Marc C.; Rigling, Marina; Blumenthal, Patrik; Segatz, Valerie; Quintanilla-Belucci, Andrès; Beisel, Julia M.; Rieke-Zapp, Jörg; Schwarz, Steffen; Lachenmeier, Dirk W.; Zhang, Yanyan
The production of coffee leaf tea (Coffea arabica) in El Salvador and the influences of processing steps on non-volatile compounds and volatile aroma-active compounds were investigated. The tea was produced according to the process steps of conventional tea (Camellia sinensis) with the available possibilities on the farm. Influencing factors were the leaf type (old, young, yellow, shoots), processing (blending, cutting, rolling, freezing, steaming), drying (sun drying, oven drying, roasting) and fermentation (wild, yeast, Lactobacillus). Subsequently, the samples were analysed for the maximum levels of caffeine, chlorogenic acid, and epigallocatechin gallate permitted by the European Commission. The caffeine content ranged between 0.37–1.33 g/100 g dry mass (DM), the chlorogenic acid was between not detectable and 9.35 g/100 g DM and epigallocatechin gallate could not be detected at all. Furthermore, water content, essential oil, ash content, total polyphenols, total catechins, organic acids, and trigonelline were determined. Gas chromatography—mass spectrometry—olfactometry and calculation of the odour activity values (OAVs) were carried out to determine the main aroma-active compounds, which are β-ionone (honey-like, OAV 132-927), decanal (citrus-like, floral, OAV 14-301), α-ionone (floral, OAV 30-100), (E,Z)-2,6-nonadienal (cucumber-like, OAV 18-256), 2,4-nonadienal (melon-like, OAV 2-18), octanal (fruity, OAV 7-23), (E)-2 nonenal (citrus-like, OAV 1-11), hexanal (grassy, OAV 1-10), and 4-heptenal (green, OAV 1-9). The data obtained in this study may help to adjust process parameters directly to consumer preferences and allow coffee farmers to earn an extra income from this by-product.
Development of rapid analytical methods for coffee quality assessment: Spectroscopy and chemometrics approach
(2024) Munyendo, Leah Masakhwe; Hitzmann, Bernd; Zhang, Yanyan
The assessment of coffee quality is based on the physical characteristics (bean quality), chemical constituents, and cup quality. Different factors, including altitude, genetics, management conditions, presence of adulterants, roasting, geographical origin, processing methods, and storage, affect the coffee quality. To meet the consumers' expectations regarding quality, the development of fast, new, and advanced analytical techniques for assessing the factors affecting coffee quality is a central aspect. Therefore, this research aimed to develop spectroscopic techniques complemented with chemometrics for evaluating the factors affecting coffee quality. The first specific objective was to investigate the ability of a deep autoencoder neural network to detect adulterants in roasted Arabica coffee and to determine a coffee’s geographical origin using near‐infrared (NIR) spectroscopy. Arabica coffee was adulterated with Robusta coffee or chicory at adulteration levels ranging from 2.5 % to 30 % in increments of 2.5 % at light, medium, and dark roast levels. Based on the results, all the samples adulterated with chicory were detectable by the autoencoder at all roast levels. For Robusta-adulterated samples, the detection was possible at adulteration levels above 7.5 % at medium and dark roasts. One can attribute the observations to potential differences in the chemical composition among the samples. Additionally, it was possible to differentiate coffee samples from different geographical origins. As a continuation of the first objective, the potential of NIR spectroscopy to quantify Robusta coffee or chicory in roasted Arabica coffee using different regression models constructed from the linear discriminant analysis (LDA) or principal component analysis (PCA) features was investigated. In addition, two classification methods (k-nearest neighbor regression (KNR) and LDA) were used. The regression models derived from LDA-extracted features exhibited better accuracies than those derived from PCA-extracted features. The two feature extraction methods exhibit differences in their working principle. PCA focuses on identifying the direction of maximum variance regardless of the adulteration levels. In contrast, LDA identifies the feature subspace that optimizes the separability of the classes (adulteration levels) and minimizes the variance within the class. Therefore, LDA extracted the features better than PCA, explaining the better performance of the regression models constructed from its features. The models provided satisfactory results with the coefficient of determination (R2) values above 0.92 for both the adulterants, indicating their efficiency in quantifying Robusta coffee or chicory in roasted Arabica coffee. For the classification methods, the LDA model performed better than KNR. Another focus of this doctoral research was to develop analytical tools based on Raman and NIR spectroscopy for real-time monitoring of the coffee roasting process by predicting chemical changes in coffee beans during roasting. Green coffee beans of Robusta and Arabica species were roasted at 240 °C for 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, and 29 minutes. Four process runs were performed for each coffee species. The spectra of the ground samples were taken using the two spectrometers and modeled by the KNR, partial least squares regression (PLSR), and multiple linear regression (MLR). All the models based on the NIR spectra provided satisfactory results for the prediction of chlorogenic acid, trigonelline, and DPPH radical scavenging activity with low relative root mean square error of prediction (pRMSEP < 9.469 %) and high R2 (> 0.916) values. Similarly, all the models based on the Raman spectra provided acceptable prediction accuracies for monitoring the dynamics of chlorogenic acid, trigonelline, and DPPH radical scavenging activity (pRMSEP < 7.849 % and R2> 0.944). In conclusion, this research proposes different approaches that would allow valuable decisions regarding coffee quality to be made quickly and efficiently. The study suggests using NIR spectroscopy to determine a coffee’s geographical origin and detect and quantify adulterants in roasted coffee. The findings reveal that the method could be a promising tool for routine coffee quality control applications in the coffee industry and other legal sectors. The study also proposes using different spectroscopic methods (NIR and Raman) to monitor a coffee roasting process. One can consider the presented approaches as essential steps toward optimizing the roasting process at an industrial scale as they permit instantaneously taking significant process decisions.
Editorial: Flavor chemistry of food: Mechanism, interaction, new advances
(2023) Huang, Mingquan; Fan, Gang; Zhang, Yanyan
Novel method for the detection of adulterants in coffee and the determination of a coffee's geographical origin using near infrared spectroscopy complemented by an autoencoder
(2023) Munyendo, Leah; Njoroge, Daniel; Zhang, Yanyan; Hitzmann, Bernd
Coffee authenticity is a foundational aspect of quality when considering coffee's market value. This has become important given frequent adulteration and mislabelling for economic gains. Therefore, this research aimed to investigate the ability of a deep autoencoder neural network to detect adulterants in roasted coffee and to determine a coffee's geographical origin (roasted) using near infrared (NIR) spectroscopy. Arabica coffee was adulterated with robusta coffee or chicory at adulteration levels ranging from 2.5% to 30% in increments of 2.5% at light, medium and dark roast levels. First, the autoencoder was trained using pure arabica coffee before being used to detect the presence of adulterants in the samples. Furthermore, it was used to determine the geographical origin of coffee. All samples adulterated with chicory were detectable by the autoencoder at all roast levels. In the case of robusta‐adulterated samples, detection was possible at adulteration levels above 7.5% at medium and dark roasts. Additionally, it was possible to differentiate coffee samples from different geographical origins. PCA analysis of adulterated samples showed grouping based on the type and concentration of the adulterant. In conclusion, using an autoencoder neural network in conjunction with NIR spectroscopy could be a reliable technique to ensure coffee authenticity.
Rapid acidification and off-flavor reduction of pea protein by fermentation with lactic acid bacteria and yeasts
(2024) Zipori, Dor; Hollmann, Jana; Rigling, Marina; Zhang, Yanyan; Weiss, Agnes; Schmidt, Herbert
Pea protein is widely used as an alternative protein source in plant-based products. In the current study, we fermented pea protein to reduce off-flavor compounds, such as hexanal, and to produce a suitable fermentate for further processing. Laboratory fermentations using 5% (w/v) pea protein suspension were carried out using four selected lactic acid bacteria (LAB) strains, investigating their growth and acidification capabilities in pea protein. Rapid acidification of pea protein was achieved with Lactococcus lactis subsp. lactis strain LTH 7123. Next, this strain was co-inoculated together with either the yeasts Kluyveromyces lactis LTH 7165, Yarrowia lipolytica LTH 6056, or Kluyveromyces marxianus LTH 6039. Fermentation products of the mixed starter cultures and of the single strains were further analyzed by gas chromatography coupled with mass spectrometry to quantify selected volatile flavor compounds. Fermentation with L. lactis LTH 7123 led to an increase in compounds associated with the “beany” off-flavors of peas, including hexanal. However, significant reduction in those compounds was achieved after fermentation with Y. lipolytica LTH 6056 with or without L. lactis LTH 7123. Thus, fermentation using co-cultures of LAB and yeasts strains could prove to be a valuable method for enhancing quality attributes of pea protein-based products.
Risk assessment of caffeine and epigallocatechin gallate in coffee leaf tea
(2022) Tritsch, Nadine; Steger, Marc C.; Segatz, Valerie; Blumenthal, Patrik; Rigling, Marina; Schwarz, Steffen; Zhang, Yanyan; Franke, Heike; Lachenmeier, Dirk W.
Coffee leaf tea is prepared as an infusion of dried leaves of Coffea spp. in hot water. It is a traditional beverage in some coffee-producing countries and has been authorized in 2020 within the European Union (EU) according to its novel food regulation. This article reviews current knowledge on the safety of coffee leaf tea. From the various ingredients contained in coffee leaves, only two were highlighted as possibly hazardous to human health, namely, caffeine and epigallocatechin gallate (EGCG), with maximum limits implemented in EU legislation, which is why this article focuses on these two substances. While the caffeine content is comparable to that of roasted coffee beans and subject to strong fluctuations in relation to the age of the leaves, climate, coffee species, and variety, a maximum of 1–3 cups per day may be recommended. The EGCG content is typically absent or below the intake of 800 mg/day classified as hepatotoxic by the European Food Safety Authority (EFSA), so this compound is suggested as toxicologically uncritical. Depending on selection and processing (age of the leaves, drying, fermentation, roasting, etc.), coffee leaf tea may exhibit a wide variety of flavors, and its full potential is currently almost unexplored. As a coffee by-product, it is certainly interesting to increase the income of coffee farmers. Our review has shown that coffee leaf tea is not assumed to exhibit risks for the consumer, apart from the well-known risk of caffeine inherent to all coffee-related beverages. This conclusion is corroborated by the history of its safe use in several countries around the world.
A robust fermentation process for natural chocolate-like flavor production with Mycetinis scorodonius
(2022) Rigling, Marina; Heger, Fabienne; Graule, Maria; Liu, Zhibin; Zhang, Chen; Ni, Li; Zhang, Yanyan
Submerged fermentation of green tea with the basidiomycete Mycetinis scorodonius resulted in a pleasant chocolate-like and malty aroma, which could be a promising chocolate flavor alternative to current synthetic aroma mixtures in demand of consumer preferences towards healthy natural and ‘clean label’ ingredients. To understand the sensorial molecular base on the chocolate-like aroma formation, key aroma compounds of the fermented green tea were elucidated using a direct immersion stir bar sorptive extraction combined with gas chromatography–mass spectrometry–olfactometry (DI-SBSE-GC-MS-O) followed by semi-quantification with internal standard. Fifteen key aroma compounds were determined, the most important of which were dihydroactinidiolide (odor activity value OAV 345), isovaleraldehyde (OAV 79), and coumarin (OAV 24), which were also confirmed by a recombination study. Furthermore, effects of the fermentation parameters (medium volume, light protection, agitation rate, pH, temperature, and aeration) on the aroma profile were investigated in a lab-scale bioreactor at batch fermentation. Variation of the fermentation parameters resulted in similar sensory perception of the broth, where up-scaling in volume evoked longer growth cycles and aeration significantly boosted the concentrations yet added a green note to the overall flavor impression. All findings prove the robustness of the established fermentation process with M. scorodonius for natural chocolate-like flavor production.
Sensorial and aroma profiles of coffee by-products - coffee leaves and coffee flowers
(2023) Rigling, Marina; Steger, Marc C.; Lachenmeier, Dirk W.; Schwarz, Steffen; Zhang, Yanyan
The utilization of coffee leaves and flowers has been underestimated over the years. Both by-products can be obtained from coffee trees without adversely affecting the production of coffee beans. To gain fundamental knowledge of their sensorial and aroma profiles, it becomes essential to reintroduce them into the food chain. Accordingly, 24 different coffee leaf samples generated from diverse processing as well as 38 varied species of coffee flowers were analyzed for their sensory characteristics by descriptive analysis and liking tests, and their corresponding aroma profiles were decoded by means of gas chromatography–mass spectrometry–olfactometry. For the coffee leaves, a wide range of different flavors could be detected in the sensory evaluation. The fermented coffee leaf samples clearly showed more sweetish and fruity aroma notes compared to the intense green and vegetable aroma of the non-fermented samples. β-Ionone (honey-like), decanal (citrus-like, floral), α-ionone (floral), octanal (fruity), and hexanal (green) were identified as key volatile compounds but distributed in different ratios. In the predominant coffee flowers, hay-like, hop-like, sage-like, dried apricot-like, and honey-like impressions were identified as major aroma descriptors in addition to a basic floral note. 2-Heptanol (fruity), 2-ethylhexanol (green), nerol (floral), and geraniol (floral) were identified as representative aroma compounds. All in all, a great variety of flavors was detected from the coffee leaves and flowers, which will not only provide an insight into the potential applications for the food market (i.e., coffee leaf tea and coffee flower tea) but will also help make coffee growing more sustainable.
Using a machine learning regression approach to predict the aroma partitioning in dairy matrices
(2024) Anker, Marvin; Borsum, Christine; Zhang, Youfeng; Zhang, Yanyan; Krupitzer, Christian
Aroma partitioning in food is a challenging area of research due to the contribution of several physical and chemical factors that affect the binding and release of aroma in food matrices. The partition coefficient measured by the Kmg value refers to the partition coefficient that describes how aroma compounds distribute themselves between matrices and a gas phase, such as between different components of a food matrix and air. This study introduces a regression approach to predict the Kmg value of aroma compounds of a wide range of physicochemical properties in dairy matrices representing products of different compositions and/or processing. The approach consists of data cleaning, grouping based on the temperature of Kmg analysis, pre-processing (log transformation and normalization), and, finally, the development and evaluation of prediction models with regression methods. We compared regression analysis with linear regression (LR) to five machine-learning-based regression algorithms: Random Forest Regressor (RFR), Gradient Boosting Regression (GBR), Extreme Gradient Boosting (XGBoost, XGB), Support Vector Regression (SVR), and Artificial Neural Network Regression (NNR). Explainable AI (XAI) was used to calculate feature importance and therefore identify the features that mainly contribute to the prediction. The top three features that were identified are log P, specific gravity, and molecular weight. For the prediction of the Kmg in dairy matrices, R2 scores of up to 0.99 were reached. For 37.0 °C, which resembles the temperature of the mouth, RFR delivered the best results, and, at lower temperatures of 7.0 °C, typical for a household fridge, XGB performed best. The results from the models work as a proof of concept and show the applicability of a data-driven approach with machine learning to predict the Kmg value of aroma compounds in different dairy matrices.