QSAR modelling for predicting adsorption of Sulfonylurea herbicides in agricultural volcanic ash-derived soils

This study developed an OECD-aligned QSAR model with an explicit applicability domain to predict the adsorption of Sulfonylurea (SUs) herbicides and methabenzthiazuron in agricultural volcanic ash-derived soils (VADS), a variable-charge system poorly represented in current pesticide adsorption QSAR models. Twenty-four SUs and methabenzthiazuron were evaluated in ten VADS, generating 250 compound-VADS systems from batch adsorption-desorption experiments. The dataset was complemented by soil physicochemical characterization, adsorption kinetics, and adsorption-desorption analysis, and descriptor-based QSAR modelling using a Lamarckian Genetic Algorithm for variable selection and Ridge regression across interaction, soil-specific, and herbicide-specific edaphic scenarios, with external validation and domain of applicability (DA). Adsorption was consistently nonlinear, and the Freundlich model best described equilibrium behavior. Across the 250 systems, K_fads ranged from 0.04 to 519.50, K_fdes from 0.24 to 229.91, and the hysteresis coefficient (H) from 0.00 to 0.94, indicating strong variability in retention and reversibility. The interaction scenario provided the main predictive result: with 210 training and 40 external observations [Formula: see text] = 0.438, [Formula: see text] = 0.347, RMSE_test = 0.492, and 30/40 (75%) DAcoverage). The final QSAR interaction retained 24 descriptors, indicating that adsorption in VADS is governed by the interplay between herbicide ionisation/polarity and 3D electronic features as well as soil organo-mineral reactive domains (OM, C/N, Fe, Cu, S, and P-Olsen). Accordingly, the predictive signal arose from herbicide-soil coupling rather than from molecular or edaphic properties alone. Within its calibrated chemical and edaphic space, the QSAR model provides a mechanistically interpretable, exploratory screening tool for identifying higher-mobility SUs-VADS combinations.