De Novo Discovery of Dual Hantavirus G_c/G_n Inhibitors Using a General-Purpose Generative AI

Background: Hantaviruses cause severe zoonotic hemorrhagic fevers with limited treatment options. Targeting the viral envelope glycoproteins G_c and G_n remains a priority. This study integrates a general-purpose generative AI (DeepSeek) with molecular docking to design and optimize novel ligands aiming for dual inhibition.

Methods: Novel SMILES strings were generated using DeepSeek, filtered for Lipinski compliance and low toxicity (SwissADME), and checked for novelty via PubChem. SMILES were converted to 3D structures using the SMILES to SDF/Mol Online Converter (FYIcenter.com). Ligands were docked into Hantavirus Gc (PDB: 5LJZ) and Gn (PDB: 5OPG) using Schrödinger Glide XP, with ribavirin and favipiravir as controls. Initial docking scores were inferior to controls, prompting AI-guided SMILES refinement and re-docking.

Results: Against Gc, the best refined ligand (FYI-CoreB-2) achieved a docking score of -5.251, outperforming favipiravir (-4.262) and ribavirin (-3.596). Against Gn, FYI-CoreB-3 scored -9.220, markedly superior to both controls. Several CoreB and CoreA variants showed enhanced binding relative to reference drugs, confirming successful refinement.

Conclusion: We provide preliminary evidence that general-purpose generative AIs like DeepSeek, alongside specialized drug-design AIs, can be effectively used for de novo inhibitor design. Our workflow produced novel ligands with improved predicted binding affinity to both Hantavirus Gc and Gn, achieving dual-target inhibition and yielding promising candidates for in vitro validation.