Deep learning-optimized multi-enzyme hydrolysis for walnut antihypertensive peptides.

This study integrated deep learning with in vitro and in vivo experiment to optimize the production of antihypertensive peptides from walnut protein. A novel multi-enzyme combination method was developed using large language models (LLMs) for enzyme screening, resulting in walnut protein hydrolysates with superior angiotensin-converting enzyme (ACE) inhibitory activity. The optimized combination (W1) exhibited 77.96 % ACE inhibition rate (0.5 mg/mL), and maintained 74.93 % of its activity after simulated digestion, demonstrating gastrointestinal stability and bioavailability. Additionally, W1 showed excellent antioxidant properties with strong ABTS and DPPH free radical scavenging capabilities. In spontaneously hypertensive rats (SHRs), W1 significantly reduced systolic and diastolic blood pressure, with effects comparable to captopril but demonstrating superior long-term efficacy. W1 also induced significant changes in serum biomarkers, including marked reductions in ACE activity and Angiotensin II (Ang II) levels, increases in bradykinin (BK) and renin levels, enhanced SOD activity, and decreased MDA content, reflecting improvements in antioxidant and vascular protective effects. Moreover, molecular docking analysis identified key peptides (VIRGNARL, PSYQPTPSL, and PQYSNAPQL) that exhibited strong binding affinity with ACE through multiple hydrogen bonds. These findings provide valuable insights into the application of deep learning for optimizing functional peptide production and offer a promising dietary approach for hypertension management.