QSAR-Guided In Silico Design of Piperazinylanthranilamide Factor Xa Inhibitors: Applicability-Domain Assessment, ADMET Profiling, Molecular Docking, Molecular Dynamics, and MM/GBSA

An integrated computational workflow was developed to prioritize piperazinylanthranilamide factor Xa (FXa) inhibitor candidates. Twenty literature compounds were geometry-optimized and described using alvaDesc to construct a constrained additive multivariate adaptive regression splines model. Internal validation and an assessment of the applicability domain guided the evaluation of 20 prospective analogs. Ten selected analogs and two same-series references were docked to FXa, with apixaban as the crystallographic comparator. Thirteen initial 100 ns molecular dynamics simulations were supplemented by two independent runs for each of PAX-M09, PAX-M15, and PAX-M18. Binding energetics were assessed using molecular mechanics/generalized Born surface area (MM/GBSA), alongside in silico developability profiling. The model achieved R2 = 0.904 and an internal leave-one-out Q2 of 0.846. PAX-M09 was the strongest strict in-domain prediction, whereas PAX-M15 was borderline. Apixaban redocking yielded a root-mean-square deviation (RMSD) of 0.953 Å. Although PAX-M17 and PAX-M19 ranked highest in docking, subsequent analyses prioritized PAX-M15, PAX-M09, and PAX-M18. PAX-M15 showed the most consistent positional behavior across replicates and consistently favorable run-specific MM/GBSA estimates (−68.76, −70.11, and −69.42 kcal/mol). PAX-M09 and PAX-M18 showed greater between-run heterogeneity. Solubility, metabolic, and toxicity liabilities remained important. The results support experimental investigation of the three candidates without changing their applicability-domain classifications or establishing superior experimental potency.