Machine Learning-Driven Optimization of a Functional Seafood-Based Blended Liquor With Enhanced Sensory and Bioactive Properties.

Perinereis aibuhitensis possess favorable nutritional and function properties that support its potential as a raw material for seafood-based liquor. However, the systematic application of P. aibuhitensis and its Maillard reaction products in liquor formulation has not been reported. Therefore, we aimed to (1) optimize the formulation of P. aibuhitensis-blended liquor (PBL) using a hybrid response surface methodology (RSM)-back-propagation (BP)-genetic algorithm (GA) model; (2) evaluate its in vitro antioxidant capacity after simulated human gastrointestinal digestion; and (3) assess its anti-inflammatory potential in lipopolysaccharide-induced RAW264.7 macrophages. Single-factor experiments and a gradient-boosted decision tree model identified the key formulation variables. RSM and BP-GA were then used to optimize the formulation. The optimized PBL was evaluated for antioxidant and anti-inflammatory activities after simulated gastrointestinal digestion. The optimal formulation per 100 mL of 60% (v/v) light-flavor sorghum baijiu comprised 39.1 mL of P. aibuhitensis Maillard reaction solution, 4.0 mL of milk vetch honey, and 2.93 g of honeysuckle-wild chrysanthemum extract. The optimized PBL achieved a high sensory score, exhibited significantly greater antioxidant activity than digested vitamin C, and significantly reduced TNF-α, IL-6, and NO levels in lipopolysaccharide-induced RAW264.7 macrophages. Overall, the hybrid RSM-BP-GA model is an effective strategy for developing seafood-based blended liquors with desirable sensory and bioactive properties. Further in vivo validation, long-term stability studies, and consumer acceptance evaluations are warranted.