Ensemble multiclassification model for predicting developmental toxicity in zebrafish.

Area of Science:

• Environmental Toxicology
• Computational Chemistry
• Ecotoxicology

Background:

• Organic compounds in aquatic environments pose significant risks to aquatic organisms.
• Assessing developmental toxicity is vital for environmental safety and early ecological risk detection.
• Existing binary classification models are insufficient for accurately classifying organic compound toxicity.

Purpose of the Study:

• To develop a multiclassification model for predicting the developmental toxicity of organic compounds.
• To improve the accuracy and detail in classifying developmental toxicity compared to binary models.
• To provide a novel tool for predicting developmental toxicity in unknown and novel compounds.

Main Methods:

• Utilized the ToxCast™ Phase I chemical library and literature data.
• Developed binary and multiclassification models using algorithms like random forest, SVM, XGBoost, AdaBoost, and C5.0.
• Employed 8 types of molecular fingerprints and 5-fold cross-validation for model establishment and validation.

Main Results:

• A multiclassification ensemble model (DT-VEM) was developed using a voting method.
• The binary ensemble model achieved a balanced accuracy of 0.918; the multiclassification model achieved 0.819.
• The DT-VEM demonstrated accuracies of 0.804, 0.834, and 0.855 for specific classifications.
• XGBoost models identified key substructures and physical properties linked to developmental toxicity.

Conclusions:

• The developed multiclassification models offer a more detailed and accurate prediction of organic compound developmental toxicity.
• The DT-VEM provides a valuable tool for assessing environmental risks posed by novel compounds.
• This research enhances the capability to predict potential developmental toxicity in previously unevaluated substances.