This study investigated the effects of hydrothermal treatment (10, 20, 30, and 40 min) on the physical, milling, thermal, and microstructural properties of proso millet grains. Hydrothermal treatment modified key engineering properties, including porosity, bulk density, and flow behaviour, coefficient of friction, and energy consumption. The grain hardness increased from 19.33 to 35.56 N, resulting in reduced grain fragmentation during milling and an increase in mean particle size from 134.7 to 449.2 μm. Consequently, specific grinding energy also increased from 420 to 700 kJ/kg, indicating higher energy requirements during milling of the hydrothermally treated grains. The treatment resulted in a 5.2
Pigeon pea starch was modified by annealing (AP), ultrasonication (US) and dual modification [annealing and ultrasonication (AUS); ultrasonication and annealing (USA)] and characterized for its physicochemical, rheological and digestibility characteristics. Modification significantly influenced starch functionality, with amylose content increasing to a maximum of 33.48
Taste perception is a fundamental sensory ability that plays a central role in regulating food intake and overall energy metabolism. Among the different taste qualities, bitterness serves as an important signal for detecting potentially harmful substances and strongly influences individual food preferences. The compound 6-n-propylthiouracil (PROP) is widely used as a marker to assess bitter taste sensitivity. Individual responses to PROP vary considerably, allowing individuals to be classified into three categories according to their perceived intensity: supertasters (ST), medium-tasters (MT), and non-tasters (NT). These differences are influenced by genetic variation in the TAS2R38 bitter taste receptor, as well as by other physiological factors, and may contribute to interindividual variation in bitter taste perception. In the present study, we specifically focused on the two extreme groups, supertasters and non-tasters, corresponding to the PAV/PAV and AVI/AVI genotypes, respectively. Heterozygous PAV/AVI individuals were not considered in the comparative analysis in order to obtain more homogeneous groups. A probabilistic dual-layer statistical physics model was employed to characterize the adsorption behavior of PROP at effective TAS2R38 interaction sites and to estimate relevant physicochemical parameters. This approach provides a quantitative framework for investigating differences in PROP taste responses between the two extreme genotype groups at the receptor-interaction level. The three complexes formed between PROP and the bitter receptor sites (TAS2R38, PAV, and AVI) have been analyzed using molecular docking and the molecular dynamics simulations.
Pickering emulsions stabilized by edible or natural particles have broad applications in food and pharmaceutical fields, yet suitable stabilizers for water-in-oil (W/O) systems remain far less abundant than those for oil-in-water (O/W) emulsions. Asiatic acid (AA) is a pentacyclic triterpenoid with diverse physiological activities and amphiphilic characteristics; however, its interfacial properties remain poorly understood. In this study, AA was modified through antisolvent precipitation (Anti-AA) to generate self-stabilized W/O emulsions. The antisolvent treatment effectively reduced particle size, induced rod-like morphology, and enhanced surface wettability. Increasing Anti-AA concentration (c = 0.5–3 wt
Pre-consumer food-processing side streams are heterogeneous solids in which wetting, release, internal and external mass transfer, and chemical change can overlap. This structured critical review examines when machine learning (ML) adds predictive value beyond empirical response-surface and physics-based kinetic models, and when apparent accuracy is insufficient for mechanistic or scale-up claims. A focused evidence appraisal identified 11 primary ML studies in the quantitative set: ten addressed extraction and one addressed enzymatic hydrolysis. None directly learned concurrent extraction-degradation kinetics or externally validated an integrated physics-informed neural network. Most studies used 15–90 designed or compiled cases, internal case-level splits, and no independent source-batch test. All 11 studies reported at least one favorable ML prediction or optimization result; however, the small, heterogeneous, predominantly within-design evidence base precluded pooled effect estimation and is compatible with publication and selective-reporting bias. The biophysical framework therefore separates phase-specific solid and liquid quantities, defines initial and boundary conditions, introduces mass-transfer Biot, Fourier, and degradation Damkohler numbers, and distinguishes prediction from parameter identification. Modern deep-learning architectures are positioned according to data structure rather than assumed superiority, while hybrid models are treated as testable parameter-learning or residual-correction strategies. A fully specified simulated example illustrates the local nature of a marginal stopping rule and propagates parameter uncertainty to the decision. Current evidence supports ML mainly as a within-domain surrogate and optimization aid; independent batch validation, calibrated uncertainty, chemically specific endpoints, and pilot-scale testing are needed before mechanistic inference or industrial transfer can be substantiated.
