Combining Generative AI, Machine Learning, and Structure-Based Drug Design to Identify Potential VEGFR1 Inhibitors

Vascular Endothelial Growth Factor Receptor 1 (VEGFR1) is a crucial target in angiogenesis, the process of forming new blood vessels, and is overexpressed in cervical cancer tumors. This overexpression promotes tumor growth and metastasis by increasing the blood supply. Targeting VEGFR1 with specific inhibitors offers a promising therapeutic approach to disrupt tumor vascularization, providing an alternative to traditional chemotherapy. In this study, we used the REINVENT generative AI model to identify potential VEGFR1 inhibitors. We fine-tuned the REINVENT prior model with known VEGFR1 inhibitors (pIC50 ≥ 8) from the ChEMBL database, generating over 2,795 novel compounds. These compounds were then filtered through a selected classification machine learning model, virtual screening with a pharmacophore model, structure-based screening, MMGBSA calculations, and ADMET predictions. To evaluate the stability and reliability of the binding interactions of these candidates, we conducted molecular dynamics (MD) simulations over 200 ns. We identified five top compounds—Compound 136, Compound 337, Compound 52, Compound 191, and Compound 202—that showed lower docking scores and binding free energies compared to the reference ligand, Sunitinib. ADMET predictions showed that, although these compounds have lower hepatotoxicity than Sunitinib, all still present elevated risk and inhibit CYP3A4 and CYP2D6, indicating potential drug–drug interactions and the need for further optimization. Duplicate 200 ns molecular dynamics (MD) simulations confirmed the dynamic stability of these complexes. The top candidate, Compound 136, exhibited minimal conformational deviation (RMSD 1.8–2.4 Å), restrained active-site residue fluctuations (RMSF < 2.0) and sustained, high-occupancy hydrogen bonds (>90%) with critical residues (GLU 878 and ASP 1040). Trajectory-derived MM-GBSA free energy calculations further verified favorable binding driven primarily by van der Waals and lipophilic interactions. This research highlights the potential of generative AI to accelerate drug discovery. The identified compounds show promise as VEGFR1 inhibitors and deserve further investigation. However, synthesizing and testing these compounds in vitro and in vivo is essential to confirm their therapeutic potential.