TPPS4 inhibits PEDV by stabilizing viral RNA G-quadruplex and promoting ER stress: a transfer-learning-driven...

Porcine epidemic diarrhea virus (PEDV) is a devastating enteric pathogen that causes substantial economic losses in the global swine industry. While PEDV inhibitors offer a promising alternative to compensate for vaccine evasion caused by viral mutations, their development is bottlenecked by poorly defined antiviral targets and limited compound libraries. Here, we developed a transfer learning framework to accelerate the discovery of anti-PEDV agents by leveraging data from human-associated coronaviruses. Transfer learning-based prediction identified Hemin as a promising PEDV inhibitor, while its analog TPPS4 exhibited potent antiviral activity with EC₅₀ values of 0.85 and 2.86 μM in Vero and LLC-PK1 cells, respectively. Notably, in vivo oral TPPS4 treatment lowered intestinal PEDV loads and doubled the piglet survival rate. Mechanistically, TPPS4 suppresses viral replication by stabilizing highly conserved G-quadruplex structures within the viral ORF1ab gene and simultaneously modulating host endoplasmic reticulum (ER) stress. This study demonstrates that transfer learning driven by human drug data facilitates the discovery of veterinary agents with diverse mechanisms of action, offering a novel paradigm for the development of other new veterinary therapeutics.

Veterinary antiviral discovery is hampered by limited bioactivity data and poorly defined targets. To address this, we developed a cross-species transfer learning pipeline to predict anti-PEDV agents. We identified TPPS4 as a potent anti-PEDV compound in vitro and in vivo, which exerts antiviral activity by stabilizing viral G-quadruplex structures and inducing host ER stress. This work establishes a workflow from computational screening to in vivo efficacy validation, demonstrating that cross-species transfer learning can accelerate veterinary antiviral discovery.