Ultrasound-assisted, machine learning-optimized extraction of silver carp viscera fish oil and microencapsulation via complex coacervation

Silver carp viscera, a lipid‑rich byproduct typically discarded, poses environmental pollution and resource wastage. Utilizing this byproduct for fish oil extraction offers a sustainable alternative; however, conventional enzymatic methods are inefficient, labor‑intensive to optimize, and yield oil susceptible to rapid oxidation. For the first time, this study developed an integrated ultrasound pretreatment and artificial neural network-genetic algorithm (ANN‑GA) strategy to maximize fish oil extraction efficiency, followed by microencapsulation in gelatin-hexametaphosphate complex coacervation microcapsules (GSM). The ANN demonstrated excellent predictive accuracy (R 2  = 0.99). Under GA-optimized conditions (ultrasonic power: 150  W, ultrasound pretreatment: 20.75  min, protease dosage: 6889 U/g protein, hydrolysis time: 1.75  h), extraction efficiency reached 79.44 %, producing oil with 46.23 % unsaturated fatty acids. Further optimization of encapsulation parameters increased GSM encapsulation efficiency (EE) to 93.33 %, surpassing gelatin‑only microcapsules (GM, 84.22 %). FT‑IR analysis indicated that electrostatic interactions induced a more ordered protein secondary structure in GSM, with α-helix and β-sheet contents rising by 5.54 % and 6.09 %, respectively. Consequently, GSM exhibited enhanced thermal stability ( T max  = 98.1 °C vs. 66.2 °C for GM), storage stability (EE unchanged for 30 days at 4 and 25°C), and oxidative stability (peroxide value approximately one-third that of GM after 10 days at 60 °C). In vitro digestion revealed GSM’s controlled release capability, liberating only 9.3 and 56 % of encapsulated fish oil during the gastric and intestinal phases, respectively. Sensory evaluation indicated significantly reduced fatty and fishy odors. Overall, this work provides a theoretical foundation for sustainable production of high‑quality freshwater fish oil.