Publicly available QSPR models for environmental media persistence

Area of Science:

• Environmental Chemistry
• Computational Toxicology
• Regulatory Science

Background:

• European Regulations mandate Persistence, Bioaccumulation and Toxicity (PBT) assessments for chemicals in environmental media.
• Existing in silico prediction tools for chemical persistence have limitations, including narrow applicability domains and a lack of medium-specific models.
• Accurate prediction of chemical persistence is crucial for environmental risk screening and prioritization.

Purpose of the Study:

• To develop and validate novel in silico models for predicting chemical persistence in water, soil, and sediment.
• To create medium-specific classification models for persistent/non-persistent (P/nP) compounds.
• To provide publicly available datasets and models for enhanced chemical risk assessment.

Main Methods:

• Compiled a dataset of 1579 unique compounds with experimental dissipation half-life values across environmental media.
• Utilized ISIDA fragment descriptors and machine learning algorithms (Support Vector Regression, Random Forest, Naïve Bayesian) for model training.
• Trained binary classification models to discriminate between persistent (P) and non-persistent (nP) compounds based on REACH half-life thresholds.

Main Results:

• Developed classification models for sediment, water, and soil with satisfactory performance (BAext: sediment=0.91, water=0.77, soil=0.76).
• The sediment model demonstrated the highest predictive accuracy.
• The soil model showed lower detection rates for persistent compounds (Sn_ext=0.50), potentially due to data source discrepancies.

Conclusions:

• The developed in silico models offer valuable tools for predicting chemical persistence in environmental media.
• Medium-specific models, particularly for sediment and water, show promising performance for PBT assessments.
• The study highlights the need for standardized half-life measurements and makes valuable data and models accessible to the scientific community.