Machine learning-assisted identification and validation of NRP1 inhibitors through molecular docking and dynamics simulations

Neuropilin-1 (NRP1) is a key mediator of tumor metastasis and progression by controlling cancer cell migration, angiogenesis, and tumor immune responses. As a result, NRP1 has recently gained considerable attention as a promising druggable target in cancer therapy. However, there are currently no FDA-approved therapeutics that inhibit NRP1, underscoring the pressing need to identify potent therapeutic candidates. Herein, a hybrid computational workflow integrating machine learning (ML), docking predictions, and molecular dynamics (MD) simulations was utilized for hunting potent NRP1 inhibitors from the NCI database. The anticipated active compounds were subsequently docked within the NRP1 active site employing docking predictions. Upon the docking findings, the top-ranking compounds were introduced to MD simulations throughout 250 ns, followed by binding energy (Δ G binding ) computations by the MM-GBSA approach. Upon the estimated Δ G binding throughout 250 ns MD simulations, three compounds —namely NCI704332, NCI717568, and NCI674566— manifested promising binding affinities with Δ G binding values of −36.7, −32.5, and −31.9 kcal/mol, respectively, relative to HRG/Arg-1 (calc. −30.2 kcal/mol). Post-MD analyses revealed the good stability of the identified NCI compounds within the NRP1 active site throughout 250 ns MD simulations. Additionally, ADME and drug-likeness assessments revealed favorable pharmacokinetic and oral bioavailability profiles for the identified NCI compounds. Ultimately, density functional theory computations were executed for the identified NCI compounds, demonstrating their high chemical reactivity. Collectively, NCI704332, NCI717568, and NCI674566 were recognized as potent candidates for in-vitro assays and further development as NRP1 inhibitors.