Developmental toxicity: artificial intelligence-powered assessments.

Area of Science:

• Toxicology and Pharmacology
• Computational Biology and Bioinformatics
• Drug Development and Regulatory Science

Background:

• Comprehensive toxicity testing for prenatal drug exposure is mandated by regulatory agencies.
• Assessing developmental toxicity, including drug-induced adverse effects in pregnant women and fetuses, presents significant challenges.
• Defining developmental toxicity endpoints and analyzing large public datasets remain complex.

Purpose of the Study:

• To provide an overview of big data resources and data-driven models for predicting developmental toxicity endpoints.
• To highlight emerging interpretable AI models for analyzing complex toxicity data.
• To propose a framework for evaluating chemical-induced developmental toxicity using AI.

Main Methods:

• Review of major big data resources relevant to developmental toxicity.
• Analysis of data-driven models, including artificial intelligence (AI) approaches, for toxicity prediction.
• Focus on interpretable AI models integrating multimodal data and domain knowledge.

Main Results:

• AI approaches are critical for analyzing high-dimensional data to uncover relationships between chemical exposures and developmental risks.
• Emerging interpretable AI models can reveal toxic mechanisms underlying complex developmental toxicity endpoints.
• A framework leveraging multiple interpretable models can comprehensively evaluate chemical-induced developmental toxicity.

Conclusions:

• AI offers powerful tools for analyzing complex datasets in developmental toxicity assessment.
• Interpretable AI models are key to understanding mechanisms of drug-induced developmental toxicity.
• A framework integrating interpretable AI can enhance the evaluation of prenatal drug safety.