AI/ML-based computational models for toxicity prediction

Area of Science:

• Computational toxicology
• Artificial Intelligence (AI)
• Machine Learning (ML)

Background:

• Increasing demand for accurate toxicity assessment and reduced animal testing drives computational model development.
• AI/ML models and online resources are crucial for modern computational toxicology research.

Purpose of the Study:

• To review computational models and data coverage for toxicity prediction and chemical safety evaluation.
• To highlight AI/ML tools for predicting various toxicity endpoints and discuss regulatory relevance.

Main Methods:

• Focus on computational models, molecular descriptors, Quantitative Structure-Activity Relationship (QSAR) models.
• Inclusion of AI/ML-based approaches, Explainable AI (XAI), and predictive methodologies.
• Analysis of data coverage, accessibility, and regulatory considerations.

Main Results:

• Computational models and AI/ML tools enable identification, prediction, and analysis of chemical toxicity across biological endpoints.
• AI/ML tools are effective for predicting neurotoxicity, hepatotoxicity, cardiotoxicity, genotoxicity, and environmental toxicity.
• Significant regulatory limitations and a lack of global conformity in chemical safety assessment were observed.

Conclusions:

• Regulatory adaptability is essential due to the rapid evolution of AI.
• Integrating AI/ML tools and interoperable frameworks can significantly advance predictive toxicology.
• Global conformity in regulatory norms is a key focus for future chemical safety evaluations.