Integrating hydrogels and machine learning to understand digestion

Licentiate thesis, 2025

Understanding the behavior of food materials during digestion is essential for advancing nutritional strategies that support individuals' health. In this study, I aim to investigate the digestive behavior of gelatin and pectin/polygalacturonic acid (PGA) gels, as model biopolymers, across different regions of the gastrointestinal tract. I also explore whether computer-assisted analysis of videos recorded during the digestion of gelatin gels can be used to determine digestion and physiological conditions.

I focused on the gastric digestion of gelatin gels crosslinked with transglutaminase (TGase), showing their stability under gastric conditions and demonstrated that convolutional neural networks (CNNs) combined with multilayer perceptrons (MLPs) have the potential to predict the degree of hydrolysis (DH) and classify digestive environments from visual features alone. To elaborate further, I examined how chemical modifications, e.g., methylation and hydrolysis, influenced the structure and rheological properties of pectin and PGA. These changes affected gel network parameters such as mesh size, turbidity, and viscosity, which are potentially linked to digestion behavior. The fermentability of modified pectin using in vitro human colonic fermentation showed that pectin of low molar mass and pectin fed as gel produced higher total short-chain fatty acids (SCFAs) than higher molar mass or dispersed forms, highlighting the importance of matrix structure in modulating microbial activity. These findings enhance our understanding of the digestion of food materials, and show potential for the use of neural networks in probing digestion from images.