Machine Learning-Enabled Drug-Induced Toxicity Prediction.

Area of Science:

• Computational toxicology
• Pharmacology
• Drug discovery

Background:

• Unexpected drug toxicity causes 30% of development failures.
• Traditional animal testing is expensive and slow.
• Artificial intelligence (AI) and machine learning (ML) offer innovative toxicology solutions.

Purpose of the Study:

• To review ML models for 10 drug-induced toxicity categories.
• To compare predictive and interpretable ML algorithms.
• To highlight key databases and tools for toxicity prediction.

Main Methods:

• Systematic review of ML applications in toxicology.
• Analysis of 10 drug-induced toxicity categories.
• Summary of relevant databases and computational tools.

Main Results:

• Identified optimal ML models vary across toxicity domains.
• Characterized applicable predictive and interpretable ML algorithms.
• Organized key databases by function for toxicity prediction.

Conclusions:

• ML shows significant promise in advancing drug-induced toxicity prediction.
• Comparative analysis of ML models is essential for optimal application.
• Strategic use of resources can bridge prediction and mechanistic insights.