Environmental toxicity risk evaluation of nitroaromatic compounds: Machine learning driven binary/multiple...

Area of Science:

• Environmental Chemistry
• Toxicology
• Computational Chemistry

Background:

• Nitroaromatic compounds (NACs) are prevalent environmental pollutants.
• Assessing the acute oral toxicity of NACs is crucial for environmental safety.
• Existing toxicity data for NACs is often limited, necessitating predictive modeling.

Purpose of the Study:

• To develop accurate predictive models for the acute oral toxicity of nitroaromatic compounds (NACs).
• To identify structural features associated with high toxicity in NACs.
• To guide the design of safer NAC alternatives.

Main Methods:

• A dataset of 371 NACs with rat oral median lethal doses (LD50s) was curated.
• Binary and multiple classification models were built using seven machine learning algorithms and six molecular fingerprints.
• Model performance was validated using 10-fold cross-validation, an external test set, and an applicability domain assessment.

Main Results:

• The Graph-RF model achieved the highest predictive performance (AUC 0.929 training, 0.956 test).
• Information gain and substructure analysis identified key structural features influencing acute oral toxicity.
• Highly toxic compounds were identified, and structural modifications led to less toxic alternatives.

Conclusions:

• Machine learning models effectively predict NAC oral toxicity, supporting environmental risk assessment.
• The study provides insights into structure-toxicity relationships for NACs.
• This work facilitates the development of greener and safer chemical alternatives.