Multimedia environmental chemical partitioning from molecular information

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Ecotoxicology

Background:

• Assessing chemical environmental distribution is crucial for risk assessment.
• Existing methods often require extensive experimental data.
• Predicting partitioning from molecular information offers a promising alternative.

Purpose of the Study:

• To develop and validate quantitative structure-fate relationship (QSFR) models for predicting chemical environmental partitioning.
• To assess the feasibility of estimating chemical distribution directly from molecular descriptors.
• To evaluate model performance across different chemical classes and compartments.

Main Methods:

• Utilized SimpleBox 3, a Level III Fugacity model, to predict multimedia chemical partitioning for 455 chemicals.
• Developed QSFR models using support vector regression with molecular descriptors (molecular weight, constituent counts).
• Employed self-organizing maps to establish the domain of applicability and validate models using an external set of 80 chemicals.

Main Results:

• Achieved high predictive accuracy (q(2) ≥ 0.90) for air and water partitioning with class-specific QSFR models (carbon-based and heteroatom-containing).
• Class-specific models showed predictive performance for carbon-based and oxygen-containing chemicals, and those with other heteroatoms.
• Prediction errors were comparable to uncertainties in physicochemical properties and degradation rates, indicating model robustness.

Conclusions:

• QSFR models can effectively predict environmental partitioning of chemicals based on molecular structure.
• Class-specific models demonstrate the potential for accurate chemical distribution assessment.
• This approach reduces reliance on experimental data, facilitating broader environmental risk evaluations.