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

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• The n-octanol/air partition coefficient (KOA) is critical for understanding the environmental transport and exposure risks of polybrominated diphenyl ethers (PBDEs).
• Accurate prediction of KOA is essential for effective environmental monitoring and risk management of PBDEs.

Purpose of the Study:

• To develop a robust computational framework for predicting the KOA of PBDE congeners.
• To identify the key molecular descriptors influencing KOA and elucidate the mechanisms of PBDE environmental partitioning.

Main Methods:

• Utilized a super learner ensemble model integrating random forest, support vector regression, and multiple linear regression.
• Employed density functional theory (DFT) derived quantum-chemical descriptors to predict KOA for 197 PBDE congeners.
• Applied SHAP analysis to determine the influence of molecular descriptors on KOA.

Main Results:

• Achieved a high predictive accuracy (R^2 = 0.981) for log KOA within the range of 7.24-11.84.
• Identified molecular polarizability and the most negative atomic charge as primary determinants of KOA.
• Demonstrated that bromination influences KOA through dispersion and electrostatic effects.
• Reduced extrapolation error by 32%-48% compared to individual algorithms.

Conclusions:

• The developed ensemble model provides high-precision estimates of PBDE KOA.
• Mechanistic insights into PBDE environmental fate are gained, highlighting the role of specific molecular properties.
• Findings support targeted monitoring and management strategies for high-mobility PBDE congeners.