Machine learning-driven design and optimization of hydroxytyrosol-modified collagen peptide systems

• Developed an ANN-GA model to accurately predict and optimize HT loading in collagen peptide (HT-CP) mixtures • Optimized conditions yielded an experimentally confirmed HT loading rate of 55.31% • Mass spectrometry provided preliminary evidence of HT-related species associated with collagen peptide fragments • Incorporation of HT into the collagen peptide matrix increased the thermal robustness of the HT-CP mixtures • Exploratory DPPH and ABTS assays showed that HT-CP mixtures maintained strong antioxidant activity Hydroxytyrosol (HT), despite its potent antioxidant properties, is hindered by structural instability. Traditional approaches to enhance its stability using peptide-based systems are often time-consuming and labor-intensive. In this study, a hybrid artificial neural network-genetic algorithm (ANN-GA) model was developed to predict and optimize the binding efficiency between HT and collagen peptides (CP), as well as to evaluate the resultant functional properties. The ANN model demonstrated superior predictive performance (R 2 = 0.96) compared to the response surface methodology (RSM, R 2 = 0.91). Global optimization using the GA algorithm yielded an experimentally validated HT binding rate of 55.31% in the optimized HT-CP complex. Preliminary mass spectrometry analysis confirmed the association between HT and CP under the experimental conditions. The optimized HT-CP system exhibited significantly enhanced thermal stability ( T max = 115.03°C vs. 97.96°C for HT) and preliminary antioxidant assays indicated improved activity (IC 50 = 0.034 mg/mL for DPPH and 0.020 mg/mL for ABTS). These results highlight the effectiveness of machine learning-based optimization in developing HT-modified peptide systems with enhanced stability and bioactivity.