Ensemble multiclassification model for aquatic toxicity of organic compounds

Area of Science:

• Environmental Toxicology
• Computational Chemistry
• Ecotoxicology

Background:

• Organic compounds pose significant environmental pollution risks, impacting aquatic organisms.
• Traditional toxicity testing is time-consuming and costly.
• Existing binary classification models lack accuracy for acute toxicity prediction.

Purpose of the Study:

• To develop and validate a multiclassification model for predicting the acute toxicity of organic compounds to aquatic organisms.
• To enhance the accuracy of toxicity classification beyond traditional methods.
• To establish a predictive tool for hazard assessment of environmental pollutants.

Main Methods:

• Utilized median lethal concentrations for 373 organic compounds from ECOTOX and EAT5 datasets.
• Classified compounds into four categories based on European Economic Community criteria.
• Employed machine learning algorithms (Random Forest, SVM, XGBoost, AdaBoost, C5.0) with eight molecular fingerprints to build base models.
• Developed an ensemble model using a voting method with 100 repetitions and fivefold cross-validation, including external validation.

Main Results:

• The best base classifier, ExtendFP-C5.0, achieved 87.30% accuracy, 87.32% sensitivity, and 95.76% specificity in external validation.
• The voting ensemble model demonstrated superior performance with 96.92% accuracy, 96.93% sensitivity, and 98.97% specificity.
• The ensemble model significantly outperformed previously reported studies in accuracy.

Conclusions:

• The developed ensemble model provides a highly accurate method for multiclassification of organic compound acute toxicity.
• This approach offers a more efficient and reliable alternative to traditional experimental methods for toxicity assessment.
• The study facilitates improved hazard classification and prediction for organic pollutants in aquatic environments.