MARSplines-Based QSAR Modeling of Asundexian Derivatives as Factor XIa Inhibitors

Factor XIa (FXIa) has emerged as an attractive molecular target for the development of anticoagulant agents. The aim of this study was to develop a quantitative structure–activity relationship (QSAR) model describing the inhibitory activity of structurally related asundexian derivatives against FXIa. The analyzed dataset comprised asundexian and 34 derivatives with experimentally reported IC₅₀ values, which were converted to pIC₅₀. Three-dimensional molecular structures were prepared in HyperChem and geometry-optimized using the MM+ force field followed by the semi-empirical PM3 method. Molecular descriptors were subsequently calculated using Dragon. Following preprocessing and variable selection, seven descriptors (Eta_sh_x, ITH, H4m, F04[N–O], Mor24u, Gu, and Mor03s) were used to construct a nonlinear QSAR model with Multivariate Adaptive Regression Splines (MARSplines) implemented in TIBCO Statistica 13.3. The resulting model showed a high goodness of fit (R² = 0.951) and strong internal cross-validation performance (Q²LOO = 0.923), with MAELOO = 0.038 and RMSELOO = 0.051. Four of the seven selected variables were three-dimensional descriptors, suggesting that molecular geometry and spatial atomic distribution contribute relevant information to the activity pattern within this congeneric series. The developed model provides an interpretable computational framework for structure–activity analysis of asundexian derivatives and may support the prioritization of structurally related FXIa inhibitors for further experimental investigation.