PepGate: A Dual-Path Diffusion Framework for ACE Inhibitory Peptide De Novo Design

Abstract

Hypertension remains a global health burden, driving demand for effective therapies. Short bioactive peptides offer high specificity and low toxicity for antihypertensive treatment, but their discovery traditionally relies on low-throughput extraction from natural proteins. Here we introduce PepGate, a gated dual-path discrete diffusion framework for de novo ACE inhibitory peptide design. By fine-tuning a protein language model into a short-peptide-specific foundation model (PepGPT), we developed an integrated pipeline comprising a generative diffusion model (PepGen) and discriminators (PepClass and PepIC50). PepGate identified 30 promising peptides with a median IC50 of 4.79 μM. In vivo validation of the top candidate, YIPVPF, demonstrated sustained blood pressure reduction in SHRs (acute SBP reduction: 48 mmHg; chronic: 43 mmHg). Computational target-network analyses further suggested broader blood-pressure-related mechanisms beyond ACE inhibition. This pipeline shifts from empirical mining to precision digital design, offering a scalable tool for next-generation therapeutic peptide development.