Rational Design of Novel Isosteviol-Derived Factor Xa Inhibitors Using Integrated QSAR, Molecular Docking, Molecular...

Factor Xa (FXa) remains an important target in the development of anticoagulant and antithrombotic agents. In this study, twenty isosteviol-derived oxime ether analogs previously reported as FXa inhibitors were used to develop a predictive QSAR model. The compounds were geometry-optimized at the B3LYP/6-311++G(d,p) level, and Dragon molecular descriptors were calculated from the optimized structures. After descriptor filtering, Random Forest-based supervised preselection and correlation-based pruning were applied. Several models of increasing complexity were evaluated, including multiple linear regression, additive MARSplines, and constrained second-order MARSplines models. The final model employed four active basis functions involving R6p+, C-025, ATSC7e, and Mor31p and demonstrated excellent calibration and cross-validated predictive ability (R² = 0.929 and Q²_LOO = 0.865). Based on this model, twenty new isosteviol-derived analogs were designed and their activities were predicted after DFT optimization and descriptor calculation. Eight representative compounds were subsequently subjected to molecular docking against human factor Xa (PDB ID: 2P16), molecular dynamics simulations, MM/GBSA binding free energy calculations, and preliminary SwissADME/pkCSM profiling. Docking protocol validation yielded a redocking RMSD of 0.953 Å. Although ISV-M20 was the highest-ranked compound according to the QSAR model, subsequent receptor-based analyses identified ISV-M19, ISV-M04, and ISV-M06 as the derivatives with the most favorable combination of structural stability, persistent protein-ligand interactions, and binding free energies. The ADMET/toxicity screen further refined this prioritization: ISV-M19, ISV-M04, and ISV-M06 showed the most favorable preliminary toxicity balance among the prioritized derivatives, whereas ISV-M20 displayed additional developability liabilities, including very high lipophilicity, poor predicted solubility, P-gp substrate status, and a predicted hERG II alert. Overall, the results demonstrate that integrating interpretable QSAR modeling with receptor-based simulations and early ADMET/toxicity filtering provides a more balanced strategy for the rational design and prioritization of novel isosteviol-derived FXa inhibitors than any single computational approach alone.