Food processing is a complex chemical process that transforms the chemical composition of the raw ingredients into their final food product, whose complexity is not yet deciphered. In our modern food system, understanding the impact of food processing on both taste and health outcomes is crucial. Traditional computational models offer some insight and utility in food production; these are essentially targeted approaches specific to foods, processes and nutritional and/or sensory outcomes. Their limitation is the inability to scale, which is necessary to address the current urgent demands of precision and personalized health and sustainable food production. These multi-variate challenges requires a deeper and more comprehensive understanding of the complexity of foods and processing methods and comprise of two main research efforts; to build food composition datasets that embody this information, and then to identify and apply the relationships as solutions. Machine Learning (ML) is widely hailed by research and industry as the technology best suited to address such an enormous multivariate problem. This shared vision has already led to efforts in building the necessary datasets. As relevant to this challenge, a common hypothesis is tested across two projects in this research - There a relationship between the chemical composition of a food and its nutritive and sensory properties in the processed state.The first project develops ML models to predict the content of seven vitamins (vitamin A, B1, B2, B3, B6, B9, C) and seven minerals (Calcium, Iron, Magnesium, Phosphorus, Potassium, Sodium, Zinc) in a processed food. The ML models are trained to learn the multi-parametric transformation patterns between the compositions of the raw and cooked foods. The focus was to be able address common dietary questions of consumers about choice of food and cooking method, and the selected training data included 425 plant and animal-based foods and 5 common cooking methods (steaming, boiling, roasting, grilling and broiling). The predictive model performed 43% and 18% better than using the standard USDA retention factor model for wet heat (steaming, boiling) and dry heat (roasting, grilling, broiling) processes, respectively. The breakdown of the predictive performance by food category revealed that legumes have the best among plant-based foods and beef the best in the animal-based foods. This suggests that nutrient loss is affected by the structural composition of foods, for future research.The second project explored structure-property models that aim to decipher the complex relation between the physical shape of a molecule and its physical properties and/or the functional role of the molecule in a product formulation. The focus was the modeling of glycans (i.e., carbohydrates), which are not only abundant in food, but essential to both food production and, more importantly, human health. In the study, regression methods were used to generalize the relationships between the structure of starch (e.g., chain length and composition of protein and amylose) and a range of its properties (e.g., gelatinization temperature, time series viscosity data, gel consistency, and sensory texture) for 301 samples of rice. The results indicated that the structure-composition data is a significantly better predictor (27% more predictive accuracy) of sensory mouthfeel than the physical properties, even though the latter is typically used in experimental research. The results of these projects demonstrate the ability of ML methods to learn a variety of complex multivariate relationships. However further progress is gated by the availability of high quality and high-resolution datasets and although the analytical methods exist, the challenge is knowing the relevant dataset for a specific prediction target. This challenge is addressed by both projects, where an assessment of what could improve prediction accuracy is the basis for future areas for data collection.