Integrated in silico, in vitro, and machine learning pipeline for extraction of antioxidant peptides and lipids from fish viscera.

Fish viscera, a processing byproduct abundant in protein and oil, remains largely underutilized. This study aimed to develop a pipeline combining in silico, in vitro, and machine-learning strategies for rapid screening of antioxidant peptides from grass carp viscera. In silico analysis identified papain as the optimal enzyme for hydrolyzing viscera proteins. Antioxidant peptides and crude fish oil were co-extracted after optimal 6-h aqueous enzymatic extraction. LightGBM model (average recall 81.2%) predicted peptide activity, leading to identification of two novel antioxidant peptides (IPGLF and FGPSGF). Molecular docking and dynamics simulations confirmed their stable binding to Keap1 via hydrogen bonds, Van der Waals force and hydrophobic interactions, with key contributions from Val and Thr residues. The co-extracted crude fish oil contained 74.29% unsaturated fatty acids, mainly oleic acid (41.93%) and linoleic acid (21.14%). This study provides a theoretical and methodological foundation for efficient valorization of fish viscera and discovery of antioxidant peptides.