Generative AI-Driven Discovery and Experimental Validation of a Novel EGFR Inhibitor for the Treatment of Pancreatic Cancer

Pancreatic cancer remains a highly lethal malignancy with limited targeted therapeutic options, necessitating the development of effective EGFR-directed inhibitors. In this study, an integrated generative artificial intelligence (AI)-driven computational–experimental workflow was developed to identify a novel EGFR inhibitor, using a cannabigerol-derived scaffold as the initial seed. ConfGAN, variational autoencoders, generative adversarial networks, and reinforcement learning were employed to explore a chemical space exceeding 1.7 million compounds, followed by progressive prioritization using drug-likeness, synthetic accessibility, molecular docking, and pharmacokinetic filtering. Selected candidates were further evaluated using 200-ns molecular dynamics simulations, free-energy landscape analysis, dynamical cross-correlation analysis, and binding free-energy calculations, with the lead candidate subsequently synthesized and evaluated using biochemical and cellular assays. This workflow identified EGFPAN_EMBS as a chemically distinct EGFR inhibitor. Relative to cannabigerol, EGFPAN_EMBS exhibited improved docking affinity, structural stability, and binding free energy, supporting favorable ligand–EGFR interactions. These computational findings translated into potent biological activity: EGFPAN_EMBS demonstrated nanomolar EGFR kinase inhibitory activity, favorable binding kinetics, and strong antiproliferative activity across pancreatic cancer cell lines, with greater selectivity toward cancer cells than the reference compounds. EGFPAN_EMBS further suppressed EGFR autophosphorylation and downstream PI3K–AKT–MAPK signaling, induced apoptosis, and promoted G₁-phase cell-cycle arrest. These findings support EGFPAN_EMBS as a chemically novel EGFR inhibitor candidate with promising biochemical and cellular activity against pancreatic cancer, and demonstrate the utility of integrating generative AI with structure-based computational analysis and experimental validation for lead discovery. The compound has been deposited in the NCBI PubChem database under Compound ID 10,972,272 and Substance ID 528,216,732.