Predicting chemical hazard across taxa through machine learning.

Area of Science:

• Environmental toxicology
• Computational toxicology
• Ecotoxicology

Background:

• Predicting chemical hazards is crucial for environmental protection.
• Existing methods often lack accuracy or rely on animal testing.
• Machine learning offers a promising alternative for toxicity prediction.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting fish acute toxicity.
• To assess the impact of taxonomic and experimental factors on prediction accuracy.
• To compare model performance against animal test reproducibility.

Main Methods:

• Applied machine learning models including K-nearest neighbors, random forests, and deep neural networks.
• Utilized Read-Across Structure Activity Relationship (RASAR) models.
• Incorporated chemical, taxonomic, and experimental data for model training and validation.

Main Results:

• Achieved prediction accuracies exceeding 93%, even on noisy datasets.
• Random forests and RASAR models demonstrated the best performance.
• Models often outperformed animal test reproducibility metrics, though careful comparison is needed.

Conclusions:

• Machine learning, especially with taxonomic and experimental data, significantly improves fish acute toxicity prediction.
• The developed approach is adaptable to various chemicals, effects, and taxa.
• This method offers a more accurate and potentially reduced-animal-testing alternative for chemical hazard assessment.