Data Science in the Food Industry

Machine Learning Prediction of Leuconostoc spp. Growth Inducing Spoilage in Cooked Deli Foods Considering the Effect of Glycine and Sodium Acetate

To control spoilage by lactic acid bacteria (Leuconostoc spp.) in cooked deli food, various combinations of environmental and/or intrinsic factors have been employed based on hurdle technology. Since many factors and their combinations greatly influence Leuconostoc spp. growth, this study aimed to develop a machine learning model based on the experimentally obtained growth kinetic data using extreme gradient boosting tree algorithm to quantitatively and flexibly predict Leuconostoc spp. growth. In particular, the effects of sodium acetate (0-1.5%) and glycine (0-1.5%), which are frequently used food additives in the Japanese food industry, on the growth of Leuconostoc spp. in cooked deli foods were examined with a combination of temperature (5-25 °C) and pH (5.0-6.0) conditions. The developed machine learning model to predict the number of Leuconostoc spp. over time successfully demonstrates comparable accuracy in culture media to the conventional Baranyi model-based prediction. Furthermore, while the accuracy of the prediction by the machine learning model for cooked deli foods such as potato salad, Japanese simmered hijiki, and unohana evaluated by the proportion of relative error within the acceptable prediction range was 98%, the accuracy of the conventional Baranyi model-based prediction was 89%. The developed machine learning model successfully and flexibly predicted the growth of Leuconostoc spp. in various cooked deli foods incorporating the effect of food additives, with an accuracy comparable to or better than that of the conventional kinetic-based model.

Contribution of data acquired from spectroscopic, genomic and microbiological analyses to enhance mussels' quality assessment

In this study, a large amount of heterogeneous data (i.e., microbiological, spectral and Next Generation Sequencing data) were obtained analyzing mussels of different species and origin, to acquire a comprehensive view about the quality and safety of these products. More specifically, spectral data were collected through Fourier transform Infrared (FTIR) spectroscopy, while the overall profile of microorganisms present in these samples, affecting quality and safety of mussels throughout storage, was determined through Next Generation Sequencing (NGS) using 16S rRNA metabarcoding analysis. In parallel, conventional microbiological analysis for the estimation of culturable spoilage microorganisms (total aerobes, Pseudomonas spp., B. thermosphacta, Shewanella spp. and Enterobacteriaceae) was applied. Different machine learning algorithms, namely Partial Least Square (PLS), Support Vector Machines (SVM), k-Nearest Neighbors (kNN), Random Forest (RF) Neural Networks (NN)) were applied accordingly, to assess the potential of FTIR and NGS data to provide useful information about mussels' microbiological quality. Microbial counts ranged from 3.5 to 9.0 log CFU/g, while NGS revealed several bacterial genera such as Pseudoalteromonas, Psychrobacter, Acinetobacter, Pseudomonas, B. thermosphacta, Psychrobacter, Kistimonas, Psychrilyobacter to affect the quality of mussels, depending on the mussel species, batch and storage conditions. According to the performance metrics, the SVM algorithm in tandem with FTIR achieved the highest prediction accuracy for microbial counts in M. chilensis samples (Rsquared; 0.89, RMSE; 0,74), while in the case of predicting the abundance of microbial genera using spectroscopic data, the best performing algorithm varied by bacterial genus. Indicatively, in M. chilensis, RF, kNN and NN performed better in predicting Enterococcus, Enhydrobacterium and Pseudoalteromonas, respectively (Rsquared = 0.92, 0.93, 0.99). Associations between genomics data and specific spectral regions were further investigated, revealing certain spectral regions that are associated with mussels' quality and safety. The application of "multi-omics" in seafood supply chain can provide insightful information about mussels' quality and safety compared to the methodologies followed in current quality and safety management systems.

Seabream Quality Monitoring Throughout the Supply Chain Using a Portable Multispectral Imaging Device

Monitoring food quality throughout the supply chain in a rapid and cost-effective way allows on-time decision-making, reducing food waste, and increasing sustainability. A portable multispectral imaging sensor was used for the rapid prediction of microbiological quality of fish fillets. Seabream fillets, packaged either in aerobic or vacuum conditions, were collected from both aquaculture and retail stores, while images were also acquired both from the skin and the flesh side of the fish fillets. In parallel to image acquisition, the microbial quality was also estimated for each fish fillet. The data were used for the training of predictive artificial neural network (ANN) models for the estimation of total aerobic counts (TACs). Models were built separately for fish parts (i.e., skin, flesh) and packaging conditions and were validated using two approaches (i.e., validation with data partitioning and external validation using samples from retail stores). The performance of the ANN models for the validation set with data partitioning was similar for the data collected from the flesh (RMSE = 0.402-0.547) and the skin side (RMSE = 0.500-0.533) of the fish fillets. Similar performance also was obtained from validation of the models of the different packaging conditions (i.e., aerobic, vacuum). The prediction capability of the models combining both air and vacuum packaged samples (RMSE = 0.531) was slightly lower compared to the models trained and validated per packaging condition, individually (RMSE = 0.510, 0.516 in air and vacuum, respectively). The models tested with unknown samples (i.e., fish fillets from retail stores-external validation) showed poorer performance (RMSE = 1.061-1.414) compared to the models validated with data partitioning (RMSE = 0.402-0.547). Multispectral imaging sensor appeared to be efficient for the rapid assessment of the microbiological quality of fish fillets for all the different cases evaluated. Hence, these outcomes could be beneficial not only for the industry and food operators but also for the authorities and consumers.

Advanced detection tools in food fraud: A systematic review for holistic and rational detection method based on research and patents

Modern food chain supply management necessitates the dire need for mitigating food fraud and adulterations. This holistic review addresses different advanced detection technologies coupled with chemometrics to identify various types of adulterated foods. The data on research, patent and systematic review analyses (2018-2023) revealed both destructive and non-destructive methods to demarcate a rational approach for food fraud detection in various countries. These intricate hygiene standards and AI-based technology are also summarized for further prospective research. Chemometrics or AI-based techniques for extensive food fraud detection are demanded. A systematic assessment reveals that various methods to detect food fraud involving multiple substances need to be simple, expeditious, precise, cost-effective, eco-friendly and non-intrusive. The scrutiny resulted in 39 relevant experimental data sets answering key questions. However, additional research is necessitated for an affirmative conclusion in food fraud detection system with modern AI and machine learning approaches.

Features in visible and Fourier transform infrared spectra confronting aspects of meat quality and fraud

Rapid assessment of microbiological quality (i.e., Total Aerobic Counts, TAC) and authentication (i.e., fresh vs frozen/thawed) of meat was investigated using spectroscopic-based methods. Data were collected throughout storage experiments from different conditions. In total 526 spectra (Fourier transform infrared, FTIR) and 534 multispectral images (MSI) were acquired. Partial Least Squares (PLS) was applied to select/transform the variables. In the case of FTIR data 30 % of the initial features were used, while for MSI-based models all features were employed. Subsequently, Support Vector Machines (SVM) regression/classification models were developed and evaluated. The performance of the models was evaluated based on the external validation set. In both cases MSI-based models (Root Mean Square Error, RMSE: 0.48-1.08, Accuracy: 91-97 %) were slightly better compared to FTIR (RMSE: 0.83-1.31, Accuracy: 88-94 %). The most informative features of FTIR for the case of quality were mainly in 900-1700 cm-1, while for fraud the features were more dispersed.

Exploration of the shelf-life difference between chilled beef and pork with similar initial levels of bacterial contamination

This study compared the shelf-life of beef and pork longissimus lumborum muscles (loins) that had the same initial bacterial loads and were held under the same chilled storage conditions. To identify the underlying pathways, comparisons were conducted from the perspective of the spoilage indicators; protease/lipase activity, and the volatile organic compounds (VOC) generated over 28 d of chilled storage. The initial total viable microbial count (TVC) on Day 0 for both type of meat was 4.3 log10 CFU/g. It was found that the TVC of beef and pork did not differ throughout the total chilled storage period and both ultimately exceeded 7 log10 CFU/g after 28 d. Based on total volatile basic nitrogen (TVB-N) guidelines, pork was spoilt after 21 d of chilled storage and therefore 7 d earlier than beef. Changes in the concentration of VOC spoilage biomarkers, including 1-octen-3-ol, 1-octanol, nonanal, and others, confirmed that pork had a shorter shelf-life than beef. An important reason for the difference in shelf-life between the two types of meat was that pork had a higher protease activity, although the beef had higher levels of total lipase activity. These findings help us understand the differences in the spoilage process of raw meat from different species and explore specific measures to control the spoilage of beef or pork.

Machine Learning Approaches for Predicting Fatty Acid Classes in Popular US Snacks Using NHANES Data

In the US, people frequently snack between meals, consuming calorie-dense foods including baked goods (cakes), sweets, and desserts (ice cream) high in lipids, salt, and sugar. Monounsaturated fatty acid (MUFA) and polyunsaturated fatty acid (PUFA) are reasonably healthy; however, excessive consumption of food high in saturated fatty acid (SFA) has been related to an elevated risk of cardiovascular diseases. The National Health and Nutrition Survey (NHANES) uses a 24 h recall to collect information on people's food habits in the US. The complexity of the NHANES data necessitates using machine learning (ML) methods, a branch of data science that uses algorithms to collect large, unstructured, and structured data sets and identify correlations between the data variables. This study focused on determining the ability of ML regression models including artificial neural networks (ANNs), decision trees (DTs), k-nearest neighbors (KNNs), and support vector machines (SVMs) to assess the variability in total fat content concerning the classes (SFA, MUFA, and PUFA) of US-consumed snacks between 2017 and 2018. KNNs and DTs predicted SFA, MUFA, and PUFA with mean squared error (MSE) of 0.707, 0.489, 0.612, and 1.172, 0.846, 0.738, respectively. SVMs failed to predict the fatty acids accurately; however, ANNs performed satisfactorily. Using ensemble methods, DTs (10.635, 5.120, 7.075) showed higher error values for MSE than linear regression (LiR) (9.086, 3.698, 5.820) for SFA, MUFA, and PUFA prediction, respectively. R2 score ranged between -0.541 to 0.983 and 0.390 to 0.751 for models one and two, respectively. Extreme gradient boost (XGR), Light gradient boost (LightGBM), and random forest (RF) performed better than LiR, with RF having the lowest score for MSE in predicting all the fatty acid classes.

Microbiological Quality Estimation of Meat Using Deep CNNs on Embedded Hardware Systems

Spectroscopic sensor imaging of food samples meta-processed by deep machine learning models can be used to assess the quality of the sample. This article presents an architecture for estimating microbial populations in meat samples using multispectral imaging and deep convolutional neural networks. The deep learning models operate on embedded platforms and not offline on a separate computer or a cloud server. Different storage conditions of the meat samples were used, and various deep learning models and embedded platforms were evaluated. In addition, the hardware boards were evaluated in terms of latency, throughput, efficiency and value on different data pre-processing and imaging-type setups. The experimental results showed the advantage of the XavierNX platform in terms of latency and throughput and the advantage of Nano and RP4 in terms of efficiency and value, respectively.

Preliminary investigation into the prediction of indicators of beef spoilage using Raman and Fourier transform infrared spectroscopy

The objective of this study was to assess the potential to predict the microbial beef spoilage indicators by Raman and Fourier transform infrared (FT-IR) spectroscopies. Vacuum skin packaged (VSP) beef steaks were stored at 0 °C, 4 °C, 8 °C and under a dynamic temperature condition (0 °C ∼ 4 °C ∼ 8 °C, for 36 d). Total viable count (TVC) and total volatile basic nitrogen (TVB-N) were obtained during the storage period along with spectroscopic data. The Raman and FTIR spectra were baseline corrected, pre-processed using Savitzky-Golay smoothing and normalized. Subsequently partial least squares regression (PLSR) models of TVC and TVB-N were developed and evaluated. The root mean squared error (RMSE) ranged from 0.81 to1.59 (log CFU/g or mg/100 g) and the determination coefficient (R²) from 0.54 to 0.75. The performance of PLSR model based on data fusion (combination of Raman and FT-IR data) is better than that based on Raman spectra and similar to that of FT-IR. Overall, Raman spectroscopy, FT-IR spectroscopy, and a combination of both exhibited a potential for the prediction of the beef spoilage.

Trends and innovations in the formulation of plant-based foods

Globally, the production, distribution, sale and consumption of plant-based foods (PBFs) are on the increase due to heightened consumer awareness, a growing demand for clean label products, widespread efforts to promote and embrace sustainable practices, and ethical concerns over animal-derived counterparts. This has led to the exploration of several strategies by researchers and the food industry to develop alternative milk, cheese, meat, and egg products from various plant-based sources using technologies such as precision fermentation (PF), scaffolding, extrusion, and muscle fibre simulation. This work explores current alternative protein sources and PBFs, production trends, innovations in formulation, nutritional quality, as well as challenges restricting full utilization and other limitations. However, PBFs have several limitations which constrain their acceptance, including the beany flavour of legumes, concerns about genetically modified foods, cost, nutritional inadequacies associated micronutrient deficiencies, absence of safety regulations, and the addition of ingredients that are contrary to their intended health-promoting purpose. The review concludes that investing in the development of PBFs now, has the potential to facilitate a rapid shift to large scale consumption of sustainable and healthy diets in the near future.

Graphical Abstract

Assessment of the Microbial Spoilage and Quality of Marinated Chicken Souvlaki through Spectroscopic and Biomimetic Sensors and Data Fusion

Fourier-transform infrared spectroscopy (FT-IR), multispectral imaging (MSI), and an electronic nose (E-nose) were implemented individually and in combination in an attempt to investigate and, hence, identify the complexity of the phenomenon of spoilage in poultry. For this purpose, marinated chicken souvlaki samples were subjected to storage experiments (isothermal conditions: 0, 5, and 10 °C; dynamic temperature conditions: 12 h at 0 °C, 8 h at 5 °C, and 4 h at 10 °C) under aerobic conditions. At pre-determined intervals, samples were microbiologically analyzed for the enumeration of total viable counts (TVCs) and Pseudomonas spp., while, in parallel, FT-IR, MSI, and E-nose measurements were acquired. Quantitative models of partial least squares–Regression (PLS-R) and support vector machine–regression (SVM-R) (separately for each sensor and in combination) were developed and validated for the estimation of TVCs in marinated chicken souvlaki. Furthermore, classification models of linear discriminant analysis (LDA), linear support vector machine (LSVM), and cubic support vector machines (CSVM) that classified samples into two quality classes (non-spoiled or spoiled) were optimized and evaluated. The model performance was assessed with data obtained by six different analysts and three different batches of marinated souvlaki. Concerning the estimation of the TVCs via the PLS-R model, the most efficient prediction was obtained with spectral data from MSI (root mean squared error—RMSE: 0.998 log CFU/g), as well as with combined data from FT-IR/MSI (RMSE: 0.983 log CFU/g). From the developed SVM-R models, the predictions derived from MSI and FT-IR/MSI data accurately estimated the TVCs with RMSE values of 0.973 and 0.999 log CFU/g, respectively. For the two-class models, the combined data from the FT-IR/MSI instruments analyzed with the CSVM algorithm provided an overall accuracy of 87.5%, followed by the MSI spectral data analyzed with LSVM, with an overall accuracy of 80%. The abovementioned findings highlighted the efficacy of these non-invasive rapid methods when used individually and in combination for the assessment of spoilage in marinated chicken products regardless of the impact of the analyst, season, or batch.

Spectroscopic Data for the Rapid Assessment of Microbiological Quality of Chicken Burgers

The rapid assessment of the microbiological quality of highly perishable food commodities is of great importance. Spectroscopic data coupled with machine learning methods have been investigated intensively in recent years, because of their rapid, non-destructive, eco-friendly qualities and their potential to be used on-, in- or at-line. In the present study, the microbiological quality of chicken burgers was evaluated using Fourier transform infrared (FTIR) spectroscopy and multispectral imaging (MSI) in tandem with machine learning algorithms. Six independent batches were purchased from a food industry and stored at 0, 4, and 8 °C. At regular time intervals (specifically every 24 h), duplicate samples were subjected to microbiological analysis, FTIR measurements, and MSI sampling. The samples (n = 274) acquired during the data collection were classified into three microbiological quality groups: “satisfactory”: 4−7 log CFU/g, “acceptable”: 7−8 log CFU/g, and “unacceptable”: >8 logCFU/g. Subsequently, classification models were trained and tested (external validation) with several machine learning approaches, namely partial least squares discriminant analysis (PLSDA), support vector machine (SVM), random forest (RF), logistic regression (LR), and ordinal logistic regression (OLR). Accuracy scores were attained for the external validation, exhibiting FTIR data values in the range of 79.41−89.71%, and, for the MSI data, in the range of 74.63−85.07%. The performance of the models showed merit in terms of the microbiological quality assessment of chicken burgers.

Application of Fourier Transform Infrared (FT-IR) Spectroscopy, Multispectral Imaging (MSI) and Electronic Nose (E-Nose) for the Rapid Evaluation of the Microbiological Quality of Gilthead Sea Bream Fillets

The potential of Fourier transform infrared (FT-IR) spectroscopy, multispectral imaging (MSI), and electronic nose (E-nose) was explored in order to determine the microbiological quality of gilthead sea bream (Sparus aurata) fillets. Fish fillets were maintained at four temperatures (0, 4, 8, and 12 °C) under aerobic conditions and modified atmosphere packaging (MAP) (33% CO₂, 19% O₂, 48% N₂) for up to 330 and 773 h, respectively, for the determination of the population of total viable counts (TVC). In parallel, spectral data were acquired by means of FT-IR and MSI techniques, whereas the volatile profile of the samples was monitored using an E-nose. Thereafter, the collected data were correlated to microbiological counts to estimate the TVC during fish fillet storage. The obtained results demonstrated that the partial least squares regression (PLS-R) models developed on FT-IR data provided satisfactory performance in the estimation of TVC for both aerobic and MAP conditions, with coefficients of determination (R²) for calibration of 0.98 and 0.94, and root mean squared error of calibration (RMSE_C) values of 0.43 and 0.87 log CFU/g, respectively. However, the performance of the PLS-R models developed on MSI data was less accurate with R² values of 0.79 and 0.77, and RMSE_C values of 0.78 and 0.72 for aerobic and MAP storage, respectively. Finally, the least satisfactory performance was observed for the E-nose with the lowest R² (0.34 and 0.17) and the highest RMSE_C (1.77 and 1.43 log CFU/g) values for aerobic and MAP conditions, respectively. The results of this work confirm the effectiveness of FT-IR spectroscopy for the rapid evaluation of the microbiological quality of gilthead sea bream fillets.