基于人工智能的食品风味组学和感官组学融合新案例

在本文中，作者从6个因素入手，以经过训练的品评人员打分为结果，将风味评级划分为5个等级，提出了融合注意力机制的深度神经网络风味评价模型。模型经过训练后，预测分数与实际 ...

研究表明，红茶香气感官品质与其含有的挥发性香气物质具有较好的关联性，如Wickremashinghe等提出的Wickremashinghe-Yamanishi比值[9] 以及Owuor等提出的Owuor风味指数(FI) [ ...

本文对基于Logistic回归分析和随机森林两种白葡萄酒质量检测方法进行比较研究，选取了4898个白葡萄酒样本，通过混淆矩阵与十折交叉验证后，得出随机森林模型在测试集精确度及 ...

... 机器学习完成检测工作。因该技术具有快速、便捷、直观、低成本等优点，在果蔬产品的分级及检测中得到了广泛应用。李思广[7] 通过机器视觉拍摄杨梅图像完成预处理、灰度 ...

展望未来，随着人工智能、物联网、多组学分析等前沿科技的深度融合，卷烟检测技术将进一步向智能化、微型化、系统化方向演进。实时在线检测、大数据驱动的质量预测与 ...

在生产工艺优化中，人工智能通过机器学习模型如神经网络、随机森林优化菌种筛选与鉴定、微生物数量动态调控及风味品质监测，显著提升菌株筛选准确率和风味分析效率，克服传统 ...

此举将使人工智能与大数据技术有效融合到食品科学领域中,融合过程中可以追踪溯源食品本源,监控食品安全,分析用户口味偏好,创新个性化食谱并进行智能推荐等,使得食品 ...

数据挖掘与预测：将大量的烟叶分析数据上传到人工智能系统，然后系统建立预测模型就可以客观预测烟叶品质、烟叶特征、烟叶产量等核心指标，为烟草企业的生产规划、资源 ...

结果表明SVR模型预测性能更优，其预测集决定系数达到0.91，预测均方根误差为0.48˚Brix，证明该便携式中红外技术可用于苹果糖度的准确快速检测。李华[40]等以水蜜桃为 ...

电子鼻，又称人工嗅觉系统，是指能够人工模拟生物嗅觉系统的功能，具有智能识别能力的气敏传感器系统；是由气敏传感器技术与近年兴起的人工智能技术相结合，对生物嗅觉 ...

在电子鼻系统中,利用气体传感器阵列获取气味信息,其响应信号包含了表达气味种类与浓度的关键信息.利用预处理方法,特征提取方法及判别模型对哺乳动物嗅觉神经系统进行模拟, ...

电子鼻通过模仿哺乳动物嗅觉系统的主要构件来检测和识别复杂的气味,由于其快速,无损等优点被应用于食品生产链早期污染和缺陷检测.作为一种分析仪器,电子鼻利用传感器 ...

黄艳等利用近红外光谱与气相离子迁移谱进行主成分分析[5]，还有通过电子鼻传感器信号构建PCA分类模型[6]，虽能满足分类要求，但这些方法存在硬件成本高、检测流程复杂 ...

基于电子鼻的方法利用电子鼻传感器测量分子气味,获取高维气味数据信息作为分子特征信息,利用机器学习或深度学习算法进行气味预测.电子鼻技术应用在生活的很多方面,如 ...

卷积神经网络(CNN)技术可准确识别食品的细微特征,比如CNN能够辨识正常水果与染色 ... 除此之外,电子鼻(E-Nose)技术可模拟人鼻嗅觉,对食品气味特性检测,评估食品 ...

茶叶中茶多酚含量电子鼻技术检测模型研究[J]. 河南农业大学学报, 2012, 46(3): ... 基于卷积神经网络的茶鲜叶主要内含物的光谱快速检测方法[J]. 中国农业大学学报 ...

Three instrumental techniques, headspace-mass spectrometry (HS-MS), mid-infrared spectroscopy (MIR) and UV-visible spectrophotometry (UV-vis), have been combined to classify virgin olive oil samples based on the presence or absence of sensory defects. The reference sensory values were provided by an official taste panel. Different data fusion strategies were studied to improve the discrimination capability compared to using each instrumental technique individually. A general model was applied to discriminate high-quality non-defective olive oils (extra-virgin) and the lowest-quality olive oils considered non-edible (lampante). A specific identification of key off-flavours, such as musty, winey, fusty and rancid, was also studied. The data fusion of the three techniques improved the classification results in most of the cases. Low-level data fusion was the best strategy to discriminate musty, winey and fusty defects, using HS-MS, MIR and UV-vis, and the rancid defect using only HS-MS and MIR. The mid-level data fusion approach using partial least squares-discriminant analysis (PLS-DA) scores was found to be the best strategy for defective vs non-defective and edible vs non-edible oil discrimination. However, the data fusion did not sufficiently improve the results obtained by a single technique (HS-MS) to classify non-defective classes. These results indicate that instrumental data fusion can be useful for the identification of sensory defects in virgin olive oils.

Multiple sensor technologies including electronic nose (E-nose), electronic tongue (E-tongue), colorimeter and texture analyzer combined with chemometrics and dada fusion strategies were applied to characterize the flavor quality of traditional Chinese fermented soybean paste. Principal components analysis (PCA) was performed to divide the selected soybean pastes into three clusters which was not completely consistent with geographical regions of selected samples. Support vector machine regression (SVR) outperformed partial least squares regression (PLSR) in quantitatively predicting sensory attributes. Additionally, prediction of overall flavor of soybean paste based on data fusion of multiple sensor information, with a correlation coefficient of prediction (Rp) of 0.9636 based on SVR, was better than prediction of E-nose and E-tongue data fusion (Rp = 0.9267). This study suggested multiple sensor technologies coupled with chemometrics can be a promising tool for flavor assessment and characterization of fermented soybean paste or other food matrixes.

Codonopsis Radix (CR) is an edible food and traditional Chinese herb medicine in China. Various varieties of Codonopsis Radix have different tastes. To make the flavor of processed food stable, two kinds of electronic sensory devices, electronic nose and electronic tongue, were used to establish a discrimination model to identify the botanical origin of each sample. The optimal model built on the 88 batches of samples was selected from the models trained with all combination of two pretreatment methods and three classification methods. A comparison were performed on the models trained on the data collected by electronic nose and electronic tongue. The results showed that the model trained on the fused dataset outperformed the models trained separately on the electronic nose data and electronic tongue data. The two preprocessing approaches could improve the prediction performance of all classification methods. Classification and Regression Tree approach performed better than Partial Least Square Discriminant Analysis and Linear Discriminant Analysis in terms of accuracy. But Classification and Regression Tree tends to assign the samples of minority class to the majority class. Meanwhile, Partial Least Square Discriminant Analysis keeps a good balance between the identification requirements of all the two groups of samples. Taking all the results above, the model built using the Partial Least Square Discriminant Analysis method on the fused data after z-score was used to identify the botanical origin of Codonopsis Radix.

Different degrees of roasting result in differences in the quality and flavor of large-leaf yellow tea. The current sensory evaluation and chemical detection methods cannot meet the requirement of online differentiation of LYT roasting degree, so an accurate and comprehensive assessment method needs to be developed urgently. First, the two aroma sensing technologies were compared. Two variable screening methods and three recognition algorithms were employed to build discriminant models. The results showed that the discrimination rate of the colorimetric sensor array (CSA) in the prediction set reached 91.89 %, outperforming that of the E-nose. Subsequently, three fusion strategies were applied to improve the discrimination accuracy. The discrimination rate of the middle fusion strategy resulted in an optimal resolution of 94.59 %. The results obtained from the homologous fusion were able to evaluate the roasting degree comprehensively and accurately, which provides a new method and idea for tea aroma quality.

Sensory experiences play an important role in consumer response, purchase decision, and fidelity towards food products. Consumer studies when launching new food products must incorporate physiological response assessment to be more precise and, thus, increase their chances of success in the market. This paper introduces a novel sensory analysis system that incorporates facial emotion recognition (FER), galvanic skin response (GSR), and cardiac pulse to determine consumer acceptance of food samples. Taste and smell experiments were conducted with 120 participants recording facial images, biometric signals, and reported liking when trying a set of pleasant and unpleasant flavors and odors. Data fusion and analysis by machine learning models allow predicting the acceptance elicited by the samples. Results confirm that FER alone is not sufficient to determine consumers' acceptance. However, when combined with GSR and, to a lesser extent, with pulse signals, acceptance prediction can be improved. This research targets predicting consumer's acceptance without the continuous use of liking scores. In addition, the findings of this work may be used to explore the relationships between facial expressions and physiological reactions for non-rational decision-making when interacting with new food products.

Quantitative relations between the sensory overall quality (OQJ) of commercial single grape variety Pinot Gris and Pinot Noir wines, defined using specific sensory attributes, and the most influencing chemical components were investigated in commercial wines from different international origins. Multiple factor analysis (MFA) was applied to achieve a comprehensive map of the quality of the samples while multivariate regression models were applied to each varietal wine to determine the sensory attributes influencing OQJ the most and to understand how the combinations of the volatile compounds influenced the olfactory sensory attributes. For Pinot Gris wine, OQJ was positively correlated with sensory attributes, like "floral" aroma, "stone-fruit" flavor, "yellow" color, "caramelized" aroma, and "tropical fruit" aroma according to an Italian panel. For Pinot Noir wine, "licorice" aroma, "cloves" aroma, "fresh wood" aroma, "red fruit" flavor, "cherry" aroma, and "spicy" flavor were positively correlated with OQJ by the same panel. Important predictors for the wine quality of Pinot Gris could be characterized, but not for Pinot Noir. Additionally, sensory tests were also carried out by different panel compositions (German and Italian). Both the German and the Italian panels preferred (based on OQJ) a Pinot Gris wine from New Zealand (Gisborne), but for different perceived characteristics (fruity and aromatic notes by the Italian panel and acidity by the German panel). For Pinot Noir, different panel compositions influenced the OQJ of the wines, as the wines from Chile (with more spicy, red fruit and woody notes) were preferred by the Italian panel, while the German panels preferred the wines from Argentina (with light, subtle woody and red fruit notes). The profile of cyclic and non-cyclic proanthocyanidins was also evaluated in the two varietal wines. No clear effect of the origin was observed, but the wines from Italy (Sicily/Puglia) were separated from the rest and were characterized by percentage ratio chemical indexes (%C-4) and (%C-5) for both varieties.

Seasonings play a key role in determining sensory attributes of instant starch noodles. Controlling and improving the quality of seasoning is becoming important. In this study, five different brands along with fifteen instant starch noodles seasonings (seasoning powder, seasoning mixture sauce and the mixture of powder and sauce) were characterized by electronic nose (e-nose) and electronic tongue (e-tongue). Feature-level fusion for the integration of the signals was introduced to integrate the e-nose and e-tongue signals, aiming at improving the performances of identification and prediction models. Principal component analysis (PCA) explained over 85.00% of the total variance in e-nose data and e-tongue data, discriminated all samples. Multilayer perceptron neural networks analysis (MLPN) modeling demonstrated that the identification rate of the combined data was basically 100%. PCA, cluster analysis (CA), and MLPN proved that the classification results acquired from the combined e-nose and e-tongue data were better than individual e-nose and e-tongue result. This work demonstrated that in combination e-nose and e-tongue provided more comprehensive information about the seasonings compared to each individual e-nose and e-tongue. E-nose and e-tongue technologies hold great potential in the production, quality control, and flavor detection of instant starch noodles seasonings.

Deep learning is evolving in nutritional epidemiology to address challenges including precise nutrition and data-driven disease modeling. Fermented dairy products consumption as the implementation of specific dietary priority contributes to a lower risk of all-cause mortality, cardiovascular disease, and obesity. Various lipid types play different roles in cardiometabolic health and fermentation process changes the lipid profile in dairy products. Leveraging the power of multiple biological datasets can provide mechanistic insights into how proteins impact lipid pathways, and establish connections among fermentation-lipid biomarkers-protein. The recent leap of deep learning has been performed in food category recognition, agro-food freshness detection, and food flavor prediction and regulation. The proposed multimodal deep learning method includes four steps: (i) Forming data matrices based on data generated from different omics layers. (ii) Decomposing high-dimensional omics data according to self-attention mechanism. (iii) Constructing View Correlation Discovery Network to learn the cross-omics correlations and integrate different omics datasets. (iv) Depicting a biological network for lipid metabolism-centered quantitative multi-omics data analysis. Relying on the

Coffee cupping includes both aroma and taste, and its evaluation considers several different attributes simultaneously to define flavor quality and therefore requires complementary data from aroma and taste. This study investigates the potential and limits of a data-driven approach to describe the sensory quality of coffee using complementary analytical techniques usually available in routine quality control laboratories. Coffee flavor chemical data from 155 samples were obtained by analyzing volatile (headspace-solid-phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS)) and nonvolatile (liquid chromatography-ultraviolet/diode array detector (LC-UV/DAD)) fractions, as well as from sensory data. Chemometric tools were used to explore the data sets, select relevant features, predict sensory scores, and investigate the networks between features. A comparison of the

Brown goat milk products have gained popularity for their unique taste and flavor. The emergence of chain-reversal phenomenon makes the design and development of goat milk products gradually tend to a consumer-oriented model. However, the precise mechanism of how browning and fermentation process causes characteristics is not clear. In an effort to understand how the treatments potentially lead to certain metabolite profile changes in goat milk, comprehensive, quantitative metabolomics and peptidomics analysis of goat milk samples after browning and fermentation were undertaken. An intelligent hybrid z-score standardization-principal components algorithm-multimodal denoizing autoencoder was used for feature fusion and hidden layer fusion in high-dimensional variable space. The fermentation process significantly improved the flavor of brown goat yogurt through the tricarboxylic acid-urea-glycolysis composite pathway. Bitter peptides HPFLEWAR, PPGLPDKY, and PPPPPKK have strong interactions with both putative dipeptidyl peptidase IV and angiotensin-converting enzyme, proving that brown goat yogurt can be considered as effective provider of potential putative dipeptidyl peptidase IV and angiotensin-converting enzyme inhibitors. The level of health-promoting bioactive components and sensory contributed to consumer selection. The proposed multimodal data integrative analysis platform was applicable to explain the effect of the dynamic changes of metabolites and peptides on consumer preferences.

Objective, non-destructive grading of tobacco remains elusive because near-infrared (NIR) spectra of leaves from different quality classes are almost indistinguishable. Here we combine a large multi-label spectral data set with a single-pass pre-processing operator to resolve these subtleties. High-resolution NIR reflectance spectra were collected from 971 flue-cured samples and annotated by expert tasters for ten sensory attributes, each compressed into 7-10 calibrated grades. We introduce Polarization-Standard Filtering (PSF), an analytic fusion of modified max-min scaling, variance-aware centering, Centralization transformation and Savitzky-Golay smoothing. PSF increases inter-sample Euclidean distances sevenfold while preserving spectral envelopes, overcoming the 0.99 cosine similarity that hampers raw data. Support-vector machines trained on PSF spectra achieve 99.7 % accuracy for seven-level quality grading-75 % higher than models using unprocessed spectra and 23 % above the best conventional pipeline. Across nine additional sensory attributes, accuracy ranged from 67 % to 98 %, with a median accuracy of 85.2 %. Notably, Fragrance 98.3 %, Impurity 98.3 %, and Mellow 94.9 % also achieved high accuracies. Sliding-window occlusion localizes the most informative wavelengths (1700-1850 nm and 2080-2300 nm) to O-H, N-H and C-H combination bands, enabling chemically interpretable feature reduction to ≤ 8 % of the original channels with minimal accuracy loss. The open PSF-NIR framework transforms tobacco flavor evaluation from subjective bench tests to an inline, multi-attribute, machine-readable process.

To develop a classification model for the five flavors of Chinese medicine using advanced multi-source intelligent sensory information fusion technology. The primary aim is to investigate the feasibility of applying this model to classify and identify the flavors of various Chinese medicines effectively. We selected 122 representative Chinese medicines, each exhibiting a single distinct flavor (sour, pungent, salty, sweet, bitter), along with 14 common foods. Utilizing the nature and flavors of these decoction pieces specified in Chinese Pharmacopeia (ChP)2020 and the inherent attributes of food components, we obtained valuable data from various sensors, including the PEN3 electronic nose, ASTREE electronic tongue, and SA402B electronic tongue. We then collected single-source data matrices from these sample sensors and a multi-source data matrix that combined the data from all sensors. Using discriminant analysis (DA), principal component analysis-discriminant analysis (PCA-DA), and K-nearest neighbor algorithm (KNN) three kinds of chemometric methods were used to establish five flavors and five-category discrimination models. The results were comprehensively evaluated with the highest correct rate of the model of leave-one-out cross-validation as the index. Upon leave-one-out cross-validation, the correct judgment rate of the five flavors, five-category two-source fusion DA discrimination model (83.8%; ASTREE + SA402B) was significantly higher than the correct judgment rate of the single-source optimal DA and KNN model (73.5%; ASTREE). Following full-sample modeling, the correct judgment rate of the five flavors, five-category three-source fusion DA discrimination model (94.9%; PEN3+ASTREE+SA402B) rose substantially. This was higher than the correct judgment rate of the single-source optimal DA model (77.9%; ASTREE) and slightly higher than the two-source optimal correct judgment rate (89.7%; PEN3 + ASTREE). Compared to single-source identification, multi-source intelligent senses information fusion (MISIF) significantly improved accuracy, providing a new outlook for identifying flavor in Chinese medicine.

Currently, extra virgin olive oil, virgin olive oil and lampante olive oil are classified using physical-chemical analyses and a sensory analysis of fruitiness and defects, which is carried out by expert panels. This manual analysis is nowadays considered to be controversial and therefore analytical methodologies, which may be automated to classify these samples, are needed. In this work, we propose using an analytical platform based on two orthogonal techniques to determine the flavour components perceived in the mouth and the components contributing to the olive oils (OOs) aroma, respectively. For the former, capillary electrophoresis with ultraviolet detector (CE-UV) and high-performance liquid chromatography with UV or fluorescence detection were explored. The CE-UV analysis provided better results with the developed chemometric models (principal component analysis, linear discriminant analysis and k-nearest neighbors method). While for the latter, headspace (HS) - gas chromatography coupling with ion mobility spectrometry (GC-IMS) was selected due to the easy applicability of this technique to classify OOs. Then both techniques, CE-UV and GC-IMS, were selected to be integrated into one analytical platform. The potential of using both complementary/orthogonal techniques was demonstrated using high-level data fusion of CE-UV and GC-IMS data.

The relationships between sensory attribute and analytical measurements, performed by electronic tongue (ET) and near-infrared spectroscopy (NIRS), were investigated in order to develop a rapid method for the assessment of umami taste. Commercially available umami products and some aminoacids were submitted to sensory analysis. Results were analysed in comparison with the outcomes of analytical measurements. Multivariate exploratory analysis was performed by principal component analysis (PCA). Calibration models for prediction of the umami taste on the basis of ET and NIR signals were obtained using partial least squares (PLS) regression. Different approaches for merging data from the two different analytical instruments were considered. Both of the techniques demonstrated to provide information related with umami taste. In particular, ET signals showed the higher correlation with umami attribute. Data fusion was found to be slightly beneficial - not so significantly as to justify the coupled use of the two analytical techniques.

Reducing sugar intake is crucial for health, and odor sweetening enhances food enjoyment and quality perception. Current research relies on subjective manual sensory evaluations, which are poorly reproducible. Traditional methods also fail to capture dynamic neural responses to odor-induced sweetness. We propose an electroencephalogram local-global fusion transformer network (EEG-LGFNet) model to decode this impact objectively. Electroencephalogram data were collected from 16 subjects under different odor and sucrose stimuli. The model captures complex neural signals by integrating local and global feature extraction mechanisms. Its performance was validated across three-time windows, demonstrating efficacy over various temporal ranges. Analysis of the coefficient of determination across brain regions confirmed the importance of the frontal, central, and parietal areas of sweetness perception. The EEG-LGFNet model excelled in quantifying odor-enhanced sweetness, significantly outperforming state-of-the-art models. This research offers new insights into odor sweetening, with applications in food development, personalized nutrition, and neuroscience.

Electronic nose (E-nose), electronic tongue (E-tongue) and colorimeter combined with data fusion strategy and different machine learning algorithms (artificial neural network, ANN; extreme gradient boosting, XGBoost; random forest regression, RFR; support vector regression, SVR) were applied to quantitatively assess and predict the freshness of horse mackerel (Trachurus japonicus) during the 90-day frozen storage. The results showed that the fusion data of the E-nose, E-tongue and colorimeter could contain more information (with a total variance contribution rate of 94.734 %) than that of the independent one. ANN, RFR and XGBoost showed good performance in predicting biochemical indexes with the R

The electronic tongue (E-tongue) system has emerged as a significant innovation, aiming to replicate the complexity of human taste perception. In spite of the advancements in E-tongue technologies, two primary challenges remain to be addressed. First, evaluating the actual taste is complex due to interactions between taste and substances, such as synergistic and suppressive effects. Second, ensuring reliable outcomes in dynamic conditions, particularly when faced with high deviation error data, presents a significant challenge. The present study introduces a bioinspired artificial E-tongue system that mimics the gustatory system by integrating multiple arrays of taste sensors to emulate taste buds in the human tongue and incorporating a customized deep-learning algorithm for taste interpretation. The developed E-tongue system is capable of detecting four distinct tastes in a single drop of dietary compounds, such as saltiness, sourness, astringency, and sweetness, demonstrating notable reversibility and selectivity. The taste profiles of six different wines are obtained by the E-tongue system and demonstrated similarities in taste trends between the E-tongue system and user reviews from online, although some disparities still exist. To mitigate these disparities, a prototype-based classifier with soft voting is devised and implemented for the artificial E-tongue system. The artificial E-tongue system achieved a high classification accuracy of ∼95% in distinguishing among six different wines and ∼90% accuracy even in an environment where more than 1/3 of the data contained errors. Moreover, by harnessing the capabilities of deep learning technology, a recommendation system was demonstrated to enhance the user experience.

A comparison was made between the traditional charcoal-grilled lamb shashliks (T) and four new methods, namely electric oven heating (D), electric grill heating (L), microwave heating (W), and air fryer treatment (K). Using E-nose, E-tongue, quantitative descriptive analysis (QDA), and HS-GC-IMS and HS-SPME-GC-MS, lamb shashliks prepared using various roasting methods were characterized. Results showed that QDA, E-nose, and E-tongue could differentiate lamb shashliks with different roasting methods. A total of 43 and 79 volatile organic compounds (VOCs) were identified by HS-GC-IMS and HS-SPME-GC-MS, respectively. Unsaturated aldehydes, ketones, and esters were more prevalent in samples treated with the K and L method. As a comparison to the RF, SVM, 5-layer DNN and XGBoost models, the CNN-SVM model performed best in predicting the VOC content of lamb shashliks (accuracy rate all over 0.95) and identifying various roasting methods (accuracy rate all over 0.92).

The authenticity of salted goose products is concerning for consumers. This study describes an integrated deep-learning framework based on a generative adversarial network and combines it with data from headspace solid phase microextraction/gas chromatography-mass spectrometry, headspace gas chromatography-ion mobility spectrometry, E-nose, E-tongue, quantitative descriptive analysis, and free amino acid and 5'-nucleotide analyses to achieve reliable discrimination of four salted goose breeds. Volatile and non-volatile compounds and sensory characteristics and intelligent sensory characteristics were analyzed. A preliminary composite dataset was generated in InfoGAN and provided to several base classifiers for training. The prediction results were fused via dynamic weighting to produce an integrated model prediction. An ablation study demonstrated that ensemble learning was indispensable to improving the generalization capability of the model. The framework has an accuracy of 95%, a root mean square error (RMSE) of 0.080, a precision of 0.9450, a recall of 0.9470, and an F1-score of 0.9460.

Study on quantitative non-volatile sensometabolome of Longjing tea remains lacked. Herein, the taste and molecular features of 42 Longjing tea samples were analyzed by sensory quantitative analysis and quantitative metabolomics. A comprehensive landscape was mapped for the first time by absolute quantification of 104 non-volatiles in tea infusions using ultra-high performance liquid chromatography-mass spectrometry. Flavan-3-ols were most abundant (1051.90-1571.98 mg/L), followed by alkaloids (447.16-620.26 mg/L), amino acids (378.15-730.41 mg/L), phenolic acids (296.88-516.93 mg/L), organic acid (98.92-163.38 mg/L), flavonol glycosides (34.02-111.59 mg/L), and others. Compound epigallocatechin gallate, caffeine, theanine, quinic acid, citric acid, kaempferol-3-O-galactosylrutinoside were most predominant in each category. Tea infusions with distinct tastes (umami vs. mellow) showed chemical differences mainly in amino acids and flavonoids, with 16 compounds as key differential. Furthermore, an effective taste evaluation and discrimination model was constructed using binary logistic regression (predictive accuracy 97.6 %, umami vs. mellow), utilizing critical marker compounds kaempferol-3-O-glucosylrutinoside and aspartic acid.

This study explores selenium enrichment of milk through microbial transformation, utilizing Geotrichum candidum LG-8, isolated from traditional dairy products. A combination of electronic tongue, electronic nose, HPLC-ICP-MS, and UHPLC-QTOF-MS were used to evaluate milk quality, organic selenium types, and various components. The milk retained a weakly acidic pH post-transformation, with a flavor profile rich in sour, sweet, and umami tastes. Odor analysis revealed predominant contributions from nitrogen oxides, aromatic compounds, ketones, and organic sulfides. Selenium concentrations increased with higher initial selenium levels, with 50 μg/mL yielding concentrations of MeSeCys, SeCys2, Se(IV), SeMet, and Se(VI) at 5.15, 20.80, 1.41, 75.31 and 0.44 μg/L respectively. Significant changes were observed in the milk's amino acid profile, with metabolite identification and metabolic pathways mapping(including KEGG). The findings highlight the potential of Geotrichum candidum LG-8 for generating functional organic selenium-enriched milk, offering promising nutritional benefits in dairy production.

Free amino acids (FAAs) play a fundamental role in determining fruit quality and stress adaptation, yet their genetic regulation remains poorly understood. Through an integrated approach combining metabolomic and sensory analyses of 120 peach (Prunus persica) hybrids, we identified glutamate as a key metabolite linking FAA content to umami taste perception. By combining genome-wide association studies (GWAS) with expression quantitative trait locus (eQTL) mapping, we identified PpGGAT1 (glutamate:glyoxylate aminotransferase) and the zinc finger transcription factor PpC2H2-3 as central regulators of glutamate metabolism. Functional characterisation revealed that overexpression of PpGGAT1 led to reduced glutamate levels and diminished umami intensity, whereas PpC2H2-3 transcriptionally suppresses PpGGAT1 to enhance glutamate accumulation. Notably, elevated glutamate levels enhanced resistance to Monilinia fructicola infection, with both genes showing significant expression changes during the progression of brown rot disease. Comparative analysis further indicated that freestone cultivars exhibit superior glutamate accumulation, a trait confirmed across 100 commercial varieties. Our findings reveal a novel regulatory module, PpC2H2-3-PpGGAT1, that coordinately modulates fruit flavour quality and defence responses against pathogens. This study provides mechanistic insights into FAA regulation in fruit crops and identifies actionable molecular targets for the development of varieties with enhanced sensory attributes and disease resistance.

Improving chili pepper aroma quality is essential for industry transformation and high-value development. However, the complex volatile composition and its quantitative relationship with sensory quality remains unresolved, limiting targeted breeding of aroma-directed varieties. This study employed quantitative descriptive analysis, E-nose, and HS-SPME-GC-MS combined with chemometrics, OAV values, and machine learning to systematically analyze aroma differences between three aroma-directed (MJ7, MJ8, MJ9) and three commercial chili peppers. Aroma-directed varieties significantly outperformed traditional peppers in fruity, floral, and sweet attributes, with MJ7 achieving a floral score of 8.44 and MJ9 a fruity score of 8.16. Among 202 identified volatile components, aroma-directed peppers predominantly contained esters and ketones, while traditional varieties were alkane-rich. OPLS-DA identified characteristic compounds including β-caryophyllene and 2-methylcarbazole, with 30 key aroma compounds identified through OAV-based quantification. An Adaptive Weighted Consensus Regression (AWCR) model established quantitative relationships between key compounds and sensory attributes, showing 33.3 % improved prediction accuracy over single machine learning approaches. Feature importance analysis revealed phenylacetaldehyde as the core compound for fruity aroma and linalool for floral notes, providing precise targets for molecular breeding of aroma-directed chili peppers.

Jinhua ham, a geographical indication agriculture product in China, possesses exquisite flavor during fermentation. Three-year Jinhua ham is highly valued for superior flavor and nutrition, but current age-discrimination relies on the experience of workers, leaving high adulteration risks. In this study, Gas Chromatography-Mass Spectrometry (GC-MS) and Gas Chromatography-Ion Mobility Spectrometry (GC-IMS) were adopted to analyze the differences of flavor compounds in Jinhua ham aged for 0.5, 1, 2 and 3 years. Fifty-four aroma-active compounds were identified through OAV ≥ 1 from Jinhua ham with different ages, with the flavor changing from mushroom to citrus/grassy/buttery, then nutty, and finally to dominant coconut, peach, and floral-fatty notes. And 18 key discriminative compounds were screened out through plotted compounds content distribution. Further, 1-octen-3-ol and decanoic acid were found applicable for discriminating the age of ham based on the scatter plot and SHAP value. Then, four age-discriminating machine-learning models (LR, SVM, NB, and DT) were constructed and exhibited high accuracy, with the highest reaching 100 %. Especially, a portable ham quality evaluation circuit was constructed based on the DT classification models due to its good generalization ability and concise structure. This study provides a theoretical basis for accurately predicting ages of Jinhua ham by GC-MS with machine learning.

Flavor serves as a key quality indicator in tomato puree (TP) processing; however, conventional methods often fall short in providing rapid and accurate assessments. To address this limitation, this study integrated flavoromics with machine learning to characterize sensory transitions and volatile changes during thermal processing and to construct a predictive model for sensory quality. Through HS-SPME-GC-MS analysis, a total of 71 volatile compounds were identified. Corresponding sensory analysis revealed a progressive shift from "Freshness," "Fruity," and "Floral" to "Cooked" and "Sourness" as heat intensity increased. Among the five models evaluated, the multilayer perceptron (MLP) demonstrated superior performance (R2 > 0.99), effectively capturing nonlinear relationships between volatiles and sensory responses. Variable importance analysis identified ten key volatiles for each sensory descriptor. Moreover, external validation and aroma recombination confirmed the model's robustness and generalization capacity. These findings offer a practical framework for flavor quality prediction and real-time control in TP production.

No abstract available

To investigate the differential effects of yeast extract (YE) and clam extract (CE) on the flavor of ginger-surimi cake, this study elucidates their flavor enhancement mechanisms using a flavoromics approach integrating comprehensive two-dimensional gas chromatography-mass spectrometry (GC × GC-MS), quantitative analysis of free amino acids and nucleotides, and interpretable machine learning models. The results revealed that, in this ginger juice-flavored system, YE promoted the conversion of leucine and isoleucine into key aroma compounds, such as 2-methylbutanal (135.92 μg/kg) and 3-methylbutanal (98.81 μg/kg), via the Strecker degradation pathway, imparting nutty and roasted aromas. In contrast, CE contained high levels of free amino acids (total 194.60 mg/100 g) and nucleotides, acting synergistically to produce a high equivalent umami concentration (0.52 g MSG/100 g). Additionally, its high content of unsaturated fatty acids (e.g., eicosapentaenoic acid, EPA) led to the formation of characteristic aldehydes like (Z)-4-decenal (odor activity value, OAV = 42) through lipid oxidation in this ginger juice-flavored system, enhancing the seafood flavor and umami perception. Furthermore, Random Forest and Shapley Additive exPlanations (SHAP) combined with Support Vector Regression (SVR, SVR-SHAP) models identified key markers such as isopulegol and alanine, and uncovered a potential synergistic relationship between glutamate and docosahexaenoic acid, as well as a statistical association between alanine and 2-methylbutanal, which collectively contribute to the overall flavor profile. This study elucidates the distinct molecular-level pathways through which YE and CE enhance the flavor of ginger-surimi cake, providing a theoretical basis for the targeted flavor regulation of aquatic products.

Infant formula (IF) is a vital nutritional source for infants. However, its sensory quality deteriorates during prolonged storage, impacting consumer acceptance. This study investigated the stage-specific sensory deterioration mechanisms and the predictive modeling of IFs during storage by integrating flavoromics, fatty acid dynamics, and machine learning. Quantitative descriptive analysis identified six key attributes driving quality loss, with one-stage IF showing the earliest and most pronounced decline. Volatile compound profiling revealed aldehydes (hexanal, octanal) and ketones as dominant off-flavor markers, correlating with lipid oxidation pathways. Fatty acid analysis showed the degradation of polyunsaturated fatty acids, particularly linoleic and α-linolenic acid, as key drivers of off-flavor generation. Using only six lipid-oxidation-derived volatiles, Random Forest and XGBoost models accurately predicted stage-specific sensory scores (with R2 > 0.85), directly linking fatty acid oxidation to flavor deterioration. This work establishes lipid oxidation as predominant degradation mechanism and provides practical, data-driven tool for shelf-life optimization.

The quality of strong-flavor Baijiu (SFB) is directly determined by key flavor compounds, which are influenced by microorganisms during fermentation. This study employed flavoromics and machine learning technologies to explore the relationship between sensory attributes and flavor compounds in SFB. Initially, sweety aroma and grain aroma were determined as the core sensory attributes impacting the quality grading of SFB. Moreover, key flavor compounds influencing the sweety aroma and grain aroma of SFB were successfully predicted through machine learning classification models. Validation experiments further confirmed the core flavor compounds influencing these sensory attributes. Predictive models revealed that core microorganisms in the fermentation pit, such as Wickerhamomyces anomalus, modulate flavor compounds, thereby affecting the expression of sweety aroma and grain aroma and ultimately determining the quality of SFB. This study demonstrated the potential of machine learning in flavor research of SFB and provided valuable insights into its flavor formation mechanisms.

Guangchenpi (GCP), which is the peel of Citrus reticulata 'Chachiensis', is widely used as an herbal medicine, tea and food ingredient in southeast Asia. Prolonging its aging process results in a more pleasant flavor and increases its profitability. Through the integration of sensory evaluation with flavoromic analysis approaches, we evaluated the correlation between the flavor attributes and the profiles of the volatiles and flavonoids of GCP with various aging years. Notably, d-limonene, γ-terpinene, dimethyl anthranilate and α-phellandrene were the characteristic aroma compounds of GCP. Besides, α-phellandrene and nonanal were decisive for consumers' perception of GCP aging time due to changes of their odor activity values (OAVs). The flavor attributes of GCP tea liquid enhanced with the extension of aging time, and limonene-1,2-diol was identified as an important flavor enhancer. Combined with machine learning models, key flavor-related metabolites could be developed as efficient biomarkers for aging years to prevent GCP adulteration.

In this study, 10 popular commercial oat milk samples were analyzed for sensory quality and flavor chemistry using the Ideal Profile Method (IPM), electronic nose (E-nose), and gas chromatography-mass spectrometry (GC-MS). Based on consumer cognitive mapping of ideal products, samples were classified into "Ideal-like" and "Ideal-exceeding" categories. Ideal-like products exhibited light white appearance, pronounced oatiness, moderate sweetness and burntness, and low graininess, presenting a balanced flavor profile, whereas Ideal-exceeding samples surpassed consumer expectations in sweetness or graininess intensity, delivering stronger sensory stimulation. Furthermore, sensory differentiation among categories primarily stemmed from synergistic effects of lipid oxidation levels (e.g., 3,5-octadien-2-one) and physical stability (fiber and protein content affecting particle size distribution). This classification framework reveals that ideal sensory quality can be achieved through diverse physicochemical pathways in commercial oat milk, providing theoretical guidance for product formulation optimization and quality standardization.

Despite the growing demand for nutritious dietary options, grain-based products remain underused in the marketplace, primarily owing to their coarse texture and unpalatable taste. This paper investigates the flavor perception of mixed grains during oral processing by elucidating the formation and release mechanisms of mixed grain flavors. The perception of flavor is mediated by the synergistic interaction of multiple senses, including sight, touch, smell, hearing, and taste. These senses not only independently shape the perception of food flavor but also modulate flavor perception through cross-modal sensory interactions. Furthermore, a flavor prediction conceptual framework is develop by combining flavoromics and artificial intelligence(AI) technologies to assist in the design of higher quality mixed grain foods. The combined application of flavoromics technology and AI technology can systematic fusion of multidimensional chemical and sensory data, with the aim of providing a powerful assistant for research in the field of flavor engineering.

The fermentation of oolong tea is a critical process that determines its quality and flavor. Current fermentation control relies on tea makers’ sensory experience, which is labor-intensive and time-consuming. In this study, using Tieguanyin oolong tea as the research object, features including the tea water loss rate, aroma, image color, and texture were obtained using weight sensors, a tin oxide-type gas sensor, and a visual acquisition system. Support vector regression (SVR), random forest (RF) machine learning, and long short-term memory (LSTM) deep learning algorithms were employed to establish models for assessing the fermentation degree based on both single features and fused multi-source features, respectively. The results showed that in the test set of the fermentation degree models based on single features, the mean absolute error (MAE) ranged from 4.537 to 6.732, the root mean square error (RMSE) ranged from 5.980 to 9.416, and the coefficient of determination (R2) values varied between 0.898 and 0.959. In contrast, the data fusion models demonstrated superior performance, with the MAE reduced to 2.232–2.783, the RMSE reduced to 2.693–3.969, and R2 increased to 0.982–0.991, confirming that feature fusion enhanced characterization accuracy. Finally, the Sparrow Search Algorithm (SSA) was applied to optimize the data fusion models. After optimization, the models exhibited a MAE ranging from 1.703 to 2.078, a RMSE from 2.258 to 3.230, and R2 values between 0.988 and 0.994 on the test set. The application of the SSA further enhanced model accuracy, with the Fusion-SSA-LSTM model demonstrating the best performance. The research results enable online real-time monitoring of the fermentation degree of Tieguanyin oolong tea, which contributes to the automated production of Tieguanyin oolong tea.

In this study, NIR quantitative prediction model was established for sensory score and physicochemical components of different varieties and quality grades of Yuezhou Longjing tea. Firstly, L, a, b color factors and diffuse reflection spectral data are collected for each sample. Subsequently, the original spectrum is preprocessed. Three techniques for selecting variables, CARS, BOSS, and SPA, were utilized to extract optimal feature bands. Finally, the spectral data extracted from feature bands were fused with L, a and b color factors to build SVR and PLSR prediction models. enabling the rapid non-destructive discrimination of different varieties and grades of Yuezhou Longjing tea. The outcomes demonstrated that BOSS was the best variable selection technique for sensory score and the distinctive caffeine wavelengths, CARS, however, was the best variable selection technique for catechins distinctive wavelengths. Additionally, the middle-level data fusion-based non-linear prediction models greatly outperformed the linear prediction models. For the prediction models of sensory score, catechins, and caffeine, the relative percent deviation (RPD) values were 2.8, 1.6, and 2.6, respectively, suggesting the good predictive ability of the models. In conclusion, evaluating the quality of the five Yuezhou Longjing tea varieties using near-infrared spectroscopy and data fusion have proved as feasible.

Background: There is a continuous demand to create new, superior sensory food experiences. In the food industry, yeast-derived flavor products (YPs) are often used as ingredients in foods to create new aromas and taste qualities that are appreciated by consumers. Methods: Chicken bouillon samples containing diverse YPs were chemically and sensorially characterized using statistical multivariate analyses. The sensory evaluation was performed using quantitative descriptive analysis (QDA) by trained panelists. Thirty-four sensory attributes were scored, including odor, flavor, mouthfeel, aftertaste and afterfeel. Untargeted metabolomic profiles were obtained using stir bar sorptive extraction (SBSE) coupled to GC-MS, RPLC-MS and targeted HILIC-MS. Results: In total, 261 volatiles were detected using GC-MS, from chemical groups of predominantly aldehydes, esters, pyrazines and ketones. Random Forest (RF) modeling revealed volatiles associated with roast odor (2-ethyl-5-methyl pyrazine, 2,3,5-trimethyl-6-isopentyl pyrazine) and chicken odor (2,4-nonadienal, 2,4-decadienal, 2-acetyl furan), which could be predicted by our combined model with R2 > 0.5. In total, 2305 non-volatiles were detected for RPLC-MS and 34 for targeted HILIC-MS, where fructose-isoleucine and cyclo-leucine-proline were found to correlate with roast flavor and odor. Furthermore, a list of metabolites (glutamate, monophosphates, methionyl-leucine) was linked to umami-related flavor. This study describes a straightforward data-driven approach for studying foods with added YPs to identify flavor-impacting correlations between molecular composition and sensory perception. It also highlights limitations and preconditions for good prediction models. Overall, this study emphasizes a matrix-based approach for the prediction of food taste, which can be used to analyze foods for targeted flavor design or quality control.

This study aims to determine the quality control indices and proper storage conditions for Ethiopian Yirgacheffe (EY) coffee. An instrumental analysis by SPME-GC-MS and a consumer sensory evaluation using a CATA questionnaire with a 9-point hedonic scale were conducted. The most abundant eleven aroma compounds of EY coffee were identified through SPME-GC-MS analysis. Among the three variables examined-coffee form (beans vs. powder), storage temperature, and storage duration-storage temperature (p < 0.0001) was identified as the most influential factor affecting the sensory quality of EY coffee products. Interaction statistical analysis of instrumental and sensory data identified 2,4-bis(1,1-dimethylethyl)-phenol as the key compound negatively affecting EY coffee's aroma, while N-furfuryl pyrrole was the most positive contributor to flavor quality. These findings were strongly supported by Pearson correlation coefficients and partial least squares (PLS) regression analysis (r = 0.75). Monitoring the levels of compounds like N-furfuryl pyrrole and 2,4-bis(1,1-dimethylethyl)-phenol can help predict the aroma profile of EY coffee products, supporting quality control during processing and storage. Although the mechanisms behind the formation and degradation of these compounds are still under investigation, understanding their variations can guide better storage practices. Avoiding improper storage temperatures is recommended to maintain optimal quality, as this helps preserve flavor and extend shelf life. PRACTICAL APPLICATION: This study provides insights into the quality control and optimal storage conditions for Ethiopian Yirgacheffe coffee by combining SPME-GC-MS instrumental analysis with consumer sensory evaluation using a CATA questionnaire. The findings indicate that storage temperature significantly impacts flavor quality, with high temperatures leading to less favorable sensory attributes. Based on our results, maintaining lower storage temperatures and optimizing storage duration could help preserve positive contributor aroma compounds and enhance flavor quality. These findings offer practical guidance for the coffee industry to improve post-harvest processing and storage strategies, ensuring better consumer acceptance.

The new superior fresh apple varieties "Qin Cui" and "Qin Mi" cultivated in recent years are highly appreciated for their unique flavors, but their key aroma-active compounds have not yet been identified. This study used SPME combined with GC × GC-QTOFMS to identify 130 volatile compounds, while 28 aroma-active compounds were screened by GC-O with OAV. Based on multivariate statistical analyses, 5 compounds were further found to be the key differential compounds for distinguishing apple varieties. Sensory evaluation showed that the aroma recombination model based on quantitative data was highly consistent with the real fruit aroma profile. Through omission experiments, it was finally confirmed that methyl 2-methylbutyrate, ethyl butyrate, ethyl 2-methylbutyrate, 2-methylbutyl acetate, propyl 2-methylbutyrate, hexyl acetate, linalool, hexanal, nonanol, (E,E)-2,4-nonadienal, 1-octen-3-one, β-damascenone as the key aroma-active compounds of the two varieties. This research will provide a scientific basis for the selection of new superior apple varieties based on flavor quality guidance.

In the present work, effective methods for determining the age of sauce-flavor Baijiu by multivariate data analysis and machine learning techniques were explored. Considering the complex and dynamic flavor changes during Baijiu storage, four analytical techniques, including gas chromatography–mass spectrometry (GC–MS), gas chromatography-ion mobility spectrometry (GC-IMS), electronic nose (E-nose) and electronic tongue (E-tongue) were integrated, to build a multilayered flavor profile of Baijiu. Four types of classification models were further constructed. The fusion data strategy combined with oversampling method of synthetic minority over-sampling technique (SMOTE) and neural network, significantly enhance the accuracy (0.96) and precision (0.97) of aged Baijiu determination (ranged from 1 year to 30 years). A total of 28 important features were screened out, including furfural, 2-hexanol (GC–MS), Area 65 (GC-IMS), and bitterness (E-tongue). Furthermore, potential correlations among different data sources were discussed. The astringency (E-tongue) showed a positive correlation with ethyl lactate (GC–MS) and Area 40 (GC-IMS).

Emotion-driven food is gaining increasingly attention in the food industry; yet conventional sensory evaluations remain constrained by subjective self-reports. To enable a more comprehensive understanding of flavor perception, this study incorporated both emotional assessment and implicit neurophysiological measurement in the evaluation of different citrus flavors. A novel hierarchical spatial fusion strategy (HSFS) was proposed to integrate electroencephalography (EEG) signals with self-reported affective responses based on the customized Pleasure-Arousal-Dominance (PAD) scales. Results revealed that positive emotions were predominant subconscious responses to citrus flavor, and significant correlations were observed between emotional states and flavor acceptability scores. While the emotional labels across individuals appeared consistent, their underlying EEG activation patterns exhibit considerable variability. Under the optimal parameters (learning rate = 0.0001, batch = 64, hidden size = 640, and four attention heads and layers), the proposed HSFS-Transformer effectively integrated the comprehensive features from both global and local attention modules. This model achieved strong performance in binary and four-class emotion classification, with accuracies of 88.84 % and 82.05 %, respectively. Additionally, it demonstrated robust results with F1-scores of 86.12 % (binary) and 81.72 % (multi-class), and Cohen's Kappa values of 72.26 % and 75.21 %, confirming the reliability and consistency of the predictions. Comparative experiments further validated the architecture, highlighting its strength in capturing the complex and sparse dynamics of EEG signals. This research advanced sensory analytics by bridging neurophysiological insights and consumer feedback, offering a promising tool for precision formulation of emotion-modulating functional foods and data-driven product innovation.

BACKGROUND Flavor is a central attribute of food quality, shaping consumer preferences and market performance. Traditional evaluation methods, such as sensory panels and basic assays, are often constrained by subjectivity, low throughput, and limited scalability. With the rise of high-throughput technologies and multimodal datasets, machine learning (ML) has emerged as a promising tool for deciphering and regulating complex flavor systems. AIM OF REVIEW This review examines current flavor detection techniques and the application of ML across diverse domains. It compares supervised learning models (SVM, DT), ensemble algorithms (XGBoost, LightGBM), and deep learning approaches (CNN, ANN). This review also discusses the contribution of three major data dimensions to flavor prediction, as well as future prospects in the field. ML enables precise flavor prediction, compound screening, and real-time process control. To support these tasks, researchers have developed integrated analytical systems that combine electronic nose (E-nose), electronic tongue (E-tongue), gas chromatography-mass spectrometry (GC-MS), and gas chromatography-ion mobility spectrometry (GC-IMS). Ensemble learning and deep learning models show strong performance when handling complex, nonlinear datasets. Explainable artificial intelligence (XAI) tools such as Shapley Additive Explanations (SHAP) improve model transparency by linking predictions to underlying features. ML models further enhance both prediction accuracy and generalizability. Innovations such as attention mechanisms, graph neural networks, and digital twins support dynamic flavor modulation. ML also aids in identifying key flavor compounds and genotype-phenotype relationships, accelerating breeding and formulation. KEY SCIENTIFIC CONCEPTS OF REVIEW ML is opening up new technological avenues in flavor science, with significant potential to predict and control flavor formation mechanisms, verify product authenticity, and support the targeted design of flavor-active compounds that align with consumer expectations for sensory appeal.

The aim of this study is to provide a new perspective on the development of masking agents by examining the application of their time-series sensory profiles. The analysis of the relationship between 14 time-intensity (TI) parameters and the beany flavor masking ability of 100 flavoring materials indicate that the values of AreaInc, DurDec, and AreaDec, TI parameters related to the flavor release in the increasing and decreasing phases, were significantly higher in the top 10 masking score materials than in the bottom 10 materials. In addition to individual analysis, machine learning analysis, which can derive complex rules from large amounts of data, was performed. Machine learning-based principal component analysis and cluster analysis of the flavoring materials presented AreaInc and AreaDec as TI parameters contributing to the classification of flavor materials and their masking ability. AreaDec was suggested to be particularly important for the beany flavor masking in the two different analyses: an effective masking can be achieved by focusing on the TI profiles of flavor materials. This study proposed that time-series profiles, which are mainly used for the understanding of the sensory characteristics of foods, can be applied to the development of masking agents.