AntiLipo: A Comprehensive Database, Deep Learning-Based Prediction Model, and Computational Study of Anti-Hyperlipidemic Peptides.

Abstract

Anti-hyperlipidemic peptides (ALPs) play important roles in regulating lipid metabolism and preventing atherosclerosis, yet conventional experimental identification methods remain time-consuming. In this study, we curated 201 experimentally validated ALPs with their sequences, sources, activity data, and metabolic pathways. Network pharmacology analysis revealed that ALPs primarily participate in canonical lipid-regulatory pathways, including AMPK and PPAR signaling, fatty acid metabolism, and cholesterol metabolism, as well as adipocytokine signaling and insulin resistance-related pathways. Based on this dataset, we developed five classification models (CNN, ESM-transformer, MLP, PhyChem-transformer, and Transformer). The transformer model achieved the best overall performance (Accuracy = 0.70, MCC = 0.40), while the MLP model attained the highest AUROC (0.76). For peptide-protein bioactivity prediction, five regression models were evaluated, with the self-attention model achieving the best performance (R² = 0.61, Pearson = 0.79). Furthermore, 500 ns molecular dynamics simulations were performed for 141 peptides against seven key lipid metabolism-related protein targets, yielding 161 active peptide-target trajectories freely available for download. Finally, we constructed AntiLipo web server ( https://hwwlab.com/webserver/antilipo ), the first specialized online database for ALPs, integrating modules for peptide structure, source, activity, mechanism, and predictive model deployment.