From In Silico De Novo Generation to In Vitro Functional Validation: Discovery of Multifunctional Antithrombotic Peptides from Food-Derived Proteins via a Hybrid LSTM-GCN and Molecular Simulation Pipeline

Abstract

Cardiovascular thromboses bring heavy global health burdens, while traditional screening of food-derived antithrombotic peptides is inefficient. This work built an integrated pipeline integrating de novo peptide generation, hybrid LSTM-GCN deep learning, and molecular simulations, screening 15,000 sequences to obtain lead peptide FPGGIP. It had a binding affinity of −6.4 kcal/mol and a stable thrombin complex (RMSD = 1.42 Å), acting as a competitive thrombin inhibitor (IC50 = 14.3 μM). FPGGIP exerted ex vivo anticoagulation, suppressed vascular smooth muscle proliferation, relieved oxidative stress, recovered cell apoptosis, alleviated endothelial activation, blocked platelet aggregation, and showed low hemolysis (<5%). As a multifunctional safe peptide, it serves as a promising candidate for cardiovascular nutraceuticals, and the pipeline enables efficient peptide mining.