Structure-based development of Glypican-3-targeting peptide candidates for hepatocellular carcinoma

Hepatocellular carcinoma remains highly lethal, and extracellular GPC3 provides a tumor-associated surface target for structure-guided peptide design. This study established an entirely in silico workflow for de novo discovery and prioritization of GPC3-targeting peptide candidates using the human GPC3 structure from Protein Data Bank entry 7ZAW. W260, Y264, and M269 were selected as coordinate-resolved hotspot residues. A single 100 ns apo-GPC3 simulation was performed, and the centroid of the most populated cluster, accounting for 35.6% of the stable trajectory, was used for downstream design. RFdiffusion generated 100 peptide backbones near the hotspot surface, and ProteinMPNN generated a nonredundant peptide sequence library from the retained backbones. PepPAP was used as a sequence-level prediction model to rank candidates by its model output, which was treated as a relative computational prediction value rather than an experimentally measured binding free energy. The top 120 PepPAP-ranked sequences were refined with Rosetta FlexPepDock, generating 600 receptor-peptide poses, of which 565 satisfied the applied geometry/contact criteria. The top 30 docked peptides were evaluated using in silico solubility and toxicity predictors; 21 passed both filters. Integrated ranking identified five final priority candidates with Rosetta interface energies from - 20.459 to -18.255 REU. These peptides are computationally prioritized candidates for future experimental validation.

The online version contains supplementary material available at https://doi.org/10.1007/s13205-026-05092-4.