A Bayesian network model for predicting aquatic toxicity mode of action using two dimensional theoretical molecular...

Area of Science:

• Environmental Toxicology
• Computational Chemistry
• Predictive Modeling

Background:

• Mode of toxic action (MoA) is crucial for understanding chemical toxicity.
• Predictive MoA classification models for aquatic toxicology are underdeveloped.
• Existing datasets lack comprehensive MoA assignments for aquatic animal toxicity.

Purpose of the Study:

• To develop a Bayesian network model for classifying aquatic toxicity MoA.
• To utilize a large dataset of chemicals with known MoA assignments.
• To identify key chemical descriptors associated with aquatic toxicity MoAs.

Main Methods:

• Generated 2D theoretical chemical descriptors using the Toxicity Estimation Software Tool.
• Developed a Bayesian network model via augmented Markov blanket discovery.
• Used a dataset of 1098 chemicals with broad MoA classifications as the target.

Main Results:

• Achieved an overall model precision of 80.2% through cross-validation.
• Demonstrated high precision for Acetylcholinesterase Inhibition (AChEI) MoA (93.5%).
• Showed lower precision for the reactivity MoA (48.5%) and high accuracy for Narcosis (80.0%).

Conclusions:

• The Bayesian network model effectively classifies aquatic toxicity MoA with reasonable accuracy.
• Markov blanket analysis simplifies complex datasets by identifying key chemical descriptors.
• The computational chemistry-based model offers a valuable tool for aquatic risk assessment.