Machine learning-based QSPR modeling for predicting the n-octanol/air partition coefficient of polybrominated diphenyl ethers.

The n-octanol/air partition coefficient (K OA) governs the trans-media transport and exposure risk of polybrominated diphenyl ethers (PBDEs). This study utilizes a super learner ensemble-integrating random forest, support vector regression, and multiple linear regression-to predict K OA based on density functional theory (DFT) derived quantum-chemical descriptors. This framework estimates properties for 197 PBDE congeners within an applicability domain of log K OA 7.24-11.84. SHAP analysis identifies molecular polarizability and the most negative atomic charge as primary determinants, demonstrating that bromination modulates K OA via dispersion and electrostatic effects. The ensemble achieves a predictive accuracy of R 2 = 0.981 and reduces extrapolation error by 32%-48% compared to individual algorithms. These findings provide high-precision estimates and mechanistic insights into the environmental fate of PBDEs, supporting hierarchical strategies for monitoring and managing high-mobility congeners effectively.