نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی، دانشگاه ایلام، ایلام، ایران
2 استادیار گروه بهداشت و صنایع غذایی ، دانشکده پیرادامپزشکی ، دانشگاه ایلام ، ایران
3 گروه بهداشت و صنایع غذایی، دانشکده پیرا دامپزشکی، دانشگاه ایلام، ایلام، ایران.
4 گروه بهداشت و صنایع غذایی، دانشکده پیرا دامپزشکی، دانشگاه ایلام، ایلام، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The Maillard reaction is a chemical reaction between free amino groups in proteins and carbonyl groups of reducing sugars. The formation of a covalent bond between protein and carbohydrate, known as a protein-saccharide conjugate or conjugate compounds, improves the functional properties of proteins and is effective in developing and enhancing the flavor and color of foods. However, without precise control, there are concerns about the formation of compounds harmful to human health. Therefore, optimizing the reaction conditions to leverage its benefits and minimize harmful compounds is essential. In this research, whey protein concentrate and beta-glucan were conjugated at different temperatures, and the final Maillard reaction products were assessed using UV-visible spectrophotometry. The data were processed using Principal Component Analysis and machine learning algorithms, including Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). The KNN algorithm demonstrated superior performance, achieving a classification accuracy of 91.04%. The SVM model, employing "one-vs-one" and "one-vs-rest" strategies, attained accuracies of 87.88% and 84.85%, respectively, while the RF model yielded the lowest accuracy (77.20%). Spectral analysis confirmed that increased temperature led to a significant formation of final Maillard products. The machine learning models based on spectral data successfully enabled the precise discrimination of samples based on process temperature. In summary, this study demonstrated that the proposed approach of integrating UV-visible spectroscopy with machine learning possesses significant potential as a fast, non-destructive, and efficient method for monitoring the Maillard reaction and optimizing thermal processes in the food industry
کلیدواژهها [English]
EXTENDED ABSTRACT
The Maillard reaction represents a fundamental chemical process in food science, involving complex interactions between amino groups in proteins and carbonyl groups in reducing sugars. This reaction plays a dual role in food processing: it generates desirable sensory attributes including characteristic flavors, appealing colors, and improved functional properties, while potentially producing harmful compounds such as advanced glycation end-products when uncontrolled. Although conventional spectroscopic methods such as fluorescence and Fourier-transform infrared (FTIR) spectroscopy are used for monitoring melanoidins, and more advanced techniques like chromatography and mass spectrometry are available for mechanistic studies, these methods are often costly, time-consuming, or require complex sample preparation. Among existing techniques, UV-Vis spectroscopy is recognized as one of the primaries and most widely used methods for tracking the formation of Maillard reaction products due to its simplicity, speed, and non-destructive nature. However, interpreting data from this technique is often challenging due to overlapping absorption bands and the non-linear nature of the reaction. To address this limitation, this study introduces an innovative approach based on integrating UV-Vis spectroscopy with machine learning algorithms. The key advantage of this approach lies in leveraging the inherent speed and simplicity of UV-Vis spectroscopy while enhancing it with the predictive power of machine learning models, enabling accurate and quantitative assessment of final Maillard reaction products (melanoidins).
The experimental design involved preparing whey protein concentrate and beta-glucan conjugates (2% w/v in phosphate buffer, pH 7.0) subjected to ultrasonic-assisted Maillard reaction at precisely controlled temperatures of 60, 70, and 80°C for 12 minutes. A benchtop hyperspectral imaging system (Specam, Parto Sanat Co., Iran) operating in the 450-900 nm spectral range with 666 channels was employed for image acquisition. The system captured 162 hyperspectral images from three treatment groups with three replicates each. Conventional spectrophotometric measurements at 420 nm provided reference data for final Maillard reaction products (melanoidins). Following image preprocessing and spectral feature extraction, three machine learning algorithms—K-Nearest Neighbors (KNN), Support Vector Machine (SVM) with "one-vs-one" and "one-vs-rest" strategies, and Random Forest (RF)—were implemented for temperature-based classification of Maillard reaction progression.
Spectrophotometric analysis validated the temperature-dependent progression of the Maillard reaction, showing statistically significant increases (p < 0.05) in absorbance at 420 nm with elevated processing temperatures. Feature selection from hyperspectral data identified seven characteristic wavelengths (400.40, 401.23, 407.84, 408.67, 455.80, 573.21, and 882.45 nm) as most discriminative for classifying Maillard reaction stages. Evaluation of machine learning models revealed that the KNN algorithm achieved superior performance with 91.04% classification accuracy, effectively distinguishing samples based on their thermal processing history. The SVM classifier demonstrated robust performance with accuracies of 87.88% and 84.85% using different multiclass strategies, while the RF algorithm registered 77.20% accuracy. The high classification accuracy establishes a direct correlation between hyperspectral features and chemical changes associated with Maillard reaction progression.
This research successfully demonstrates that integration of UV-Vis spectroscopy with machine learning algorithms provides an effective, non-destructive methodology for monitoring and classifying Maillard reaction products. The identification of key spectral wavelengths associated with reaction progression offers new insights into the spectral signatures of Maillard chemistry. The outstanding performance of the KNN algorithm (91.04% accuracy) highlights the potential of this integrated approach for real-time quality control in food processing operations. This technology enables precise monitoring of thermal processing effects on Maillard reaction development, facilitating optimized process conditions that maximize desirable sensory attributes while minimizing formation of potentially harmful compounds. The method presents significant advantages over conventional analytical techniques, including non-destructive operation, rapid analysis capability, and suitability for industrial implementation, establishing a foundation for advanced process analytical technology in food manufacturing.
M.H.N; Software, Validation, Methodology, Formal analysis, Investigation, Data curation, Writing–review & editing
S.A; Conceptualization, Formal analysis, Writing-Original Draf, Data curation, Writing–review & editing
M.K; Investigation, Methodology, Data curation
All authors have read and agreed to the published version of the manuscript.
Data available on request from the authors.
This research was carried out in Ilam University, Ilam-IRAN. Therefore, the authors are thankful to Ilam University for their supporting.
The authors avoided data fabrication, falsification, plagiarism, and misconduct.
The author declares no conflict of interest.
)