Recent Advances in Machine Learning Models for Predicting Toxicity of Inorganic Nanoparticles.

Area of Science:

• Nanotechnology
• Toxicology
• Computational Chemistry

Background:

• Engineered nanoparticles (NPs) raise safety concerns due to potential cytotoxicity in biomedical and environmental applications.
• Efficient hazard and risk assessment strategies are crucial for the safe development and sustainable use of NPs.

Purpose of the Study:

• To systematically review nanotoxicity mechanisms, influencing factors, and prediction models.
• To highlight the role of machine learning (ML) in accelerating the development of nano-quantitative structure-activity relationship (nanoQSAR), physiologically based pharmacokinetic (PBPK), and meta-analysis (MA) models for cytotoxicity prediction.

Main Methods:

• Review of nanotoxicity mechanisms and classical statistical cytotoxicity prediction models (nanoQSAR, PBPK, MA).
• Focus on ML-accelerated model development, including algorithms, advantages, and schemes.
• Discussion of prediction performance, key influencing features, and relevant nanotoxicity databases.

Main Results:

• ML significantly enhances the development and predictive power of nanoQSAR, PBPK, and MA models.
• Identified key features influencing NP cytotoxicity and discussed model performance.
• Compiled important nanotoxicity databases for further research.

Conclusions:

• ML-driven models offer a powerful approach for predicting NP cytotoxicity, crucial for risk assessment.
• Further development of robust, interpretable ML models is needed to ensure NP safety in various applications.
• This review provides a comprehensive overview and highlights future directions in ML-based nanotoxicity prediction.