Recipient Organization
UNIVERSITY OF CALIFORNIA, DAVIS
410 MRAK HALL
DAVIS,CA 95616-8671
Performing Department
(N/A)
Non Technical Summary
Food processing and cooking share many common goals, such as preserving food, improving taste, and increasing food nutritional value. With changing lifestyles, there has been an undeniable increase in consumption of processed foods as well as increases in non-communicable diseases in the past several decades. This has led some to speculate that processed food consumption results in negative health consequences. However, there is a lack of comprehensive knowledge that compares similar foods that have been industrially processed or pre-processed (e.g. frozen or refrigerated products that need to be partially prepared at home) with home cooking to understand the impact of processing or preparation method on food physical properties (e.g. texture, mouthfeel), chemical properties (e.g. composition of macro- and micro-nutrients), and nutritional properties (digestibility or availability of nutrients in the human body after the food is consumed). The overall goal of this project is to develop a comprehensive assessment of the impact of processing or cooking methods on the physical, chemical, and nutritional properties of a broad range of food products. The outcomes of this project will provide a holistic understanding of the physical, chemical, and structural properties of food as a function of processing or cooking methods and their influence on the nutritional properties of foods, including protein and starch digestion and the release of micronutrients.To achieve these goals, this project integrates advances in thermal and non-thermal food processing, to compare these processing techniques with conventional (home) cooking methods to develop foods with a range of structural and chemical properties. This range of food products, including protein- and carbohydrate-rich foods, fruits, vegetables, and juices, will be analyzed using state-of-the-art approaches to assess their physicochemical characteristics and nutritional properties. The physicochemical properties of foods analyzed in this research will include texture, rheology, particle size, proximate analysis (composition), amino acid and fatty acid profiles, and salt content. The nutritional properties of foods will be measured using dynamic digestion models to quantify the release rate of macronutrients, structural breakdown, micronutrient bioaccessibility, and total macronutrient digestion. The resulting data will be analyzed using machine learning models to evaluate the roles of processing methods and the food properties on the food nutritional profile. These machine learning models will be able to classify foods into different classes (e.g. poor, low, medium, high) of nutritional properties based on food composition and processing. This information will be useful for food producers, as it will allow them to optimize their food processing and formulation to improve the nutritional properties of processed and pre-processed foods. As a result, the American food consumers will also benefit by having an increased supply of foods with better nutritional quality, as well as with knowledge in the difference in nutritional quality between processed and home cooked foods. The foundational set of knowledge developed in this project will also be able to address gaps in food classification by providing a comprehensive data set that links processing method and physicochemical properties (including composition) to food nutritional properties. Ultimately, this project seeks to improve our understanding of the impact of food processing on food nutritional quality to increase the availability and consumption of healthy food.
Animal Health Component
85%
Research Effort Categories
Basic
15%
Applied
85%
Developmental
(N/A)
Goals / Objectives
The overall goal of this project is to understand food physical, chemical, and structural properties that impact nutritional properties after industrial processing or pre-processing, novel processing, or home cooking, and to develop a data-driven framework to predict food nutritional properties that can be utilized in processing and formulation of novel foods to improve consumer health. This goal will be accomplished through the following specific objectives:(1) Identify the influence of industrial or novel processing or in-home preparation on physical, chemical, and nutritional properties of food.(2) Develop a comprehensive characterization of structural properties, chemical profile, and nutritional profile to evaluate the influence of processing on a subset of foods and realistic mixed meals.(3) Develop machine learning frameworks to identify the food properties needed to classify processed foods based on their nutritional properties (e.g. digestion behavior) that can be utilized for future food classification.
Project Methods
Objective 1: 75 foods will be utilized from three broad food groups: carbohydrate-based (8 foods), protein-based (5 foods), and fruits and vegetables (7 foods). Each food will be purchased, processed, or prepared in each of four categories: industrial processing, novel processing, pre-processed (industrial pre-processing followed by home preparation), and home cooking. The selected 20 foods are commonly consumed and meet the following criteria: (1) the industrially-processed version has been labeled ultra-processed by previous studies; (2) a food similar to the industrial version can be processed using novel technologies and; (3) similar products can be prepared in a home kitchen. These foods will have similar (macronutrient) composition but diverse structures and physicochemical properties.For each of these selected foods, the industrially (traditional) processed and the pre-processed foods will be purchased from a local supermarket, with care being taken to select brands that are available nationwide and not from local suppliers, such that the results will be broadly applicable across the country. Each food will be purchased from different lots at least in triplicate. An entire package of processed product will be thoroughly mixed and at least 3 - 5 subsamples will be taken to result in a representative sample. Foods used to prepare the home cooked products will be purchased in bulk or from a local wholesale produce retailer (for fruits and vegetables) to minimize biological variability. Each food will be prepared using standard recipes (e.g. from America's Test Kitchen or Betty Crocker cookbooks) and typical kitchen appliances (e.g. baking oven, fryer, grill, etc.). Novel processing will be conducted as described in previous studies if such novel processed foods are not commercially available using techniques such as convection-microwave baking, microwave air frying, and high pressure processing.Once the 75 foods have been prepared or purchased, their properties will be characterized. Physical properties, including texture, rheology, and the particle size and surface charge will be measured using standard methods. Chemical properties, including proximate analysis (moisture, fat, protein, ash, starch, free sugar, and dietary fiber), amino acid and fatty acid profile, salt content, acrylamide content, mineral content, and micronutrient content will be determined using standard analyses adapted to each food material. Nutritional properties will be measured using a high-throughput dynamic digestion model (peristaltic simulator) developed in the Bornhorst lab at UC Davis to quantify released macronutrients in the liquid phase of digesta, structural breakdown, micronutrient bioaccessibility, and total macronutrient digestibility (starch, protein, lipid). Twelve modules are available for simultaneous analysis in this model to facilitate high-throughput testing. All experiments will be conducted on at least in triplicate batches of foods and statistical analysis will be performed using SAS. Compositional data will be utilized to calculate the NRF9.3 value for each food, as this has been recommended by an IUFoST taskforce to differentiate processing- and formulation-induced changes in foods. An analysis of variance (ANOVA) will be utilized to determine significant differences (p < 0.05).Objective 2: Using the data generated in Objective 1, classification methods such as PCA and t-SNE will be used to reduce dimensionality of the selected foods and group foods with similar nutritional properties into clusters. We anticipate classification of these foods into 5-7 major clusters representing each of the broad food categories, where 1-2 foods will be selected from each cluster for detailed property characterization. For this subset of foods, detailed nutritional properties will be characterized using a dynamic gastric model available in the Bornhorst lab that mimics the dynamic physical and biochemical environment in the human stomach. Physical breakdown, gastric emptying, macronutrient digestion kinetics and total digestibility, and micronutrient bioaccessibility will be measured. Detailed chemical properties will be characterized using nuclear magnetic resonance (NMR) spectroscopy and infrared (IR) spectroscopy. Detailed structural properties will be characterized using reflectance confocal imaging (CSLM) and multiphoton microscopy to visualize food microstructure, and using micro-computed tomography (micro-CT) to visualize food macrostructure.Most foods are consumed as part of a mixed meal containing carbohydrates, protein, and fruit and/or vegetables. As a result, to understand the overall impact of consuming foods with different processing on nutritional outcomes, two mixed meals will be selected with foods from each of the different categories. Once the mixed meals are selected, they will be prepared using foods from each processing category. The ratios of each individual food product within the meal will be standardized across the meals to have an equivalent total carbohydrate, protein, and fat content that aligns with current dietary consumption patterns as per the National Health and Nutrition Examination Survey. For these meals, detailed nutritional characterization will be completed using dynamic digestion models (as described above for the subset of foods in Objective 2). All experiments will be conducted on triplicate batches or lots of food for each treatment and statistical analysis will be performed using SAS. ANOVA will be utilized to determine significant (p < 0.05) differences between treatments. In addition, kinetic data will be fit to appropriate empirical equations as a data reduction technique to analyze differences by comparison of parameters, as we hypothesize the kinetics will play a critical role in understanding differences in food nutritional behavior with the different processing treatments.Objective 3: First, the extensive property data collected (on the subset of individual foods) will be utilized to develop a machine learning model to use the chemical and structural data to predict the nutritional data. The input to the machine learning model will be either: NMR spectra, IR spectra, CSLM images, or micro-CT images. The detailed nutritional properties (determined in Objective 2) will be utilized to group foods into 3-4 categories in terms of their starch, protein, and lipid digestion. Multiple machine learning algorithms will be evaluated to discover the optimal model for predicting food nutritional properties based on the input data. To evaluate the machine learning models, we will randomly select 20% of the data to be held out for model validation. The remaining 80% of the data will be used for model training and testing. Model training and testing will be done through 5-fold cross-validation. The model will be validated on the 20% of the unseen data not used in testing and training.After development of the initial machine learning framework, we will develop a broader machine learning model using the quality metrics (generated in Objective 1) to classify foods based on their nutritional properties. This study will be performed in two stages. In the first stage, we will repeat the machine learning model development described above, but with the model inputs as the quality metrics generated in Objective 1 for this subset of foods. To determine the quality metrics that are most relevant, we will conduct a correlation analysis between the physical and chemical properties from Objective 1 and the chemical and structural properties from Objective 2. Then, a machine learning model will be developed with all 75 food and 8 mixed meals tested in Objectives 1 and 2 as inputs and nutritional information as outputs. We will utilize a similar approach to that described in development of the first machine learning model above for the model testing, training, and validation.