Machine learning-driven QSAR models for predicting the cytotoxicity of five common microplastics.

Area of Science:

• Environmental toxicology
• Computational toxicology
• Biomedical science

Background:

• Microplastics (MPs) pose environmental and health risks.
• Predicting MP toxicity is crucial for risk assessment.
• Machine learning (ML) offers advanced computational tools for toxicity prediction.

Purpose of the Study:

• To develop and evaluate ML-based quantitative structure-activity relationship (QSAR) models for predicting MP toxicity on BEAS-2B cells.
• To identify key features influencing MP toxicity.
• To assess the predictive performance and applicability domain of the developed models.

Main Methods:

• Six ML algorithms were employed to build QSAR models.
• Model performance was evaluated using R-squared (R2) values for training and testing datasets.
• Feature importance was determined using Embedded Feature Importance (EFI), Recursive Feature Elimination (RFE), and SHapley Additive exPlanations (SHAP).
• Williams plot analysis was used to assess model applicability domains.

Main Results:

• The extreme gradient boosting model demonstrated superior predictive performance (R2_tra = 0.9876, R2_test = 0.9286).
• All six developed models exhibited stable predictions within their applicability domains, with minimal outliers.
• Particle size was consistently identified as the most significant feature impacting MP toxicity prediction across all feature importance methods.

Conclusions:

• The developed QSAR models provide a reliable method for predicting MP toxicity.
• Particle size is a critical determinant of MP toxicity.
• These models can support preliminary environmental exposure assessments and enhance understanding of MP-related health risks.