A Stacking Ensemble Model for K_OC Prediction and Environmental Mobility Assessment of Global Neutral Chemicals

Neutral persistent, mobile, and toxic (PMT) chemicals are an emerging regulatory concern, because they can contaminate drinking water resources. The organic carbon sorption coefficient (K_OC) is widely used as a mobility indicator for neutral organic compounds (NOCs), but global mobility assessments are hampered by the scarcity of K_OC experimental data and the limited efficiency of predictive approaches. This study develops a descriptor-light stacking model (DL-SM) tailored for NOCs, which integrates 13 molecular descriptors and four tree-based learners into a multilayer perceptron. Trained on 1987 NOCs, DL-SM achieves high predictive performance (R²_test = 0.825) and outperforms five commonly used mobility assessment tools on an independent validation test. Applying DL-SM to 129,875 NOCs from 12 national and regional inventories worldwide shows that 59.28%-67.79% of NOCs possess intrinsic mobility potential. Specially, the resulting mobility score distributions are remarkably similar across inventories despite around 80% of mobile substances being unique to a single inventory. Further analysis reveals that small molecular size (e.g., nC ≤ 15, Vabc ≤ 260 ų) strongly drives mobility, while long methyl chains may attenuate the mobility-enhancing effects of hydrophilic groups. This work improves the mechanistic understanding of neutral chemical mobility and provides a robust and publicly available tool to support the identification and management of neutral PMT candidates.