New approach methodologies for risk assessment using deep learning.

Area of Science:

• Computational toxicology
• Bioinformatics
• Artificial intelligence in risk assessment

Background:

• Technological advancements and AI enable rapid processing of large biological datasets.
• Increasing biological data (3D structures, interaction networks) supports novel risk assessment methods.
• Artificial intelligence (AI) offers predictive capabilities as New Approach Methodologies (NAMs) for revolutionizing risk assessment.

Purpose of the Study:

• To develop an AI-based decision tool for chemical risk assessment.
• To utilize existing toxicity data (e.g., LD50) and predicted chemical-protein interactions.
• To support risk assessment for multiple, uncharacterized stressors, focusing on pesticides' effects on humans.

Main Methods:

• Leveraging AI for analyzing large biological and chemical toxicity datasets.
• Employing molecular docking predictions to assess chemical-protein binding affinity.
• Integrating in vivo data from literature and technical reports for validation of developed NAMs.

Main Results:

• Previous studies identified high-affinity binding of toxic chemicals to human proteins (nervous, reproductive functions).
• Identified potential sublethal interactions of neonicotinoids with bee immune system proteins.
• Established a foundation for developing an AI tool to predict toxic and sublethal effects.

Conclusions:

• AI-driven bioinformatics methodologies can significantly impact toxicity studies.
• The developed NAMs will guide experimental designs, enhancing predictability.
• These approaches promise to substantially reduce reliance on animal testing in toxicology.