Predicting In Vitro Neurotoxicity Induced by Nanoparticles Using Machine Learning.

Area of Science:

• Nanotoxicology
• Computational Toxicology
• In Vitro Assays

Background:

• Non-testing approaches are crucial for efficient nanoparticle hazard assessment.
• Machine learning (ML) shows promise in predicting nanotoxicological outcomes.
• Existing models often lack tissue specificity for neurotoxicity prediction.

Purpose of the Study:

• To develop and validate a tissue-specific ML model for predicting nanoparticle-induced neurotoxicity.
• To identify key factors influencing nanoparticle neurotoxicity in vitro.
• To establish a cost-effective and timely risk assessment tool.

Main Methods:

• Compiled a dataset of nanoparticle physicochemical properties, exposure conditions, and in vitro characteristics.
• Applied data pre-processing techniques including normalization and synthetic minority over-sampling technique (SMOTE).
• Developed a Random Forest classification model, evaluated using goodness-of-fit, robustness, and predictability metrics.

Main Results:

• Identified exposure dose, duration, toxicological assay, cell type, and zeta potential as critical predictors of neurotoxicity.
• The developed tissue-specific model demonstrated superior performance compared to non-tissue specific models.
• Information gain analysis highlighted key features driving neurotoxic predictions.

Conclusions:

• This study presents the first tissue-specific ML tool for in vitro nanoparticle neurotoxicity prediction.
• The model offers a more accurate and efficient method for assessing nanoparticle risks.
• This approach supports the development of safer nanomaterials through predictive toxicology.