Machine learning models for classification tasks related to drug safety.

Area of Science:

• Pharmacokinetics and Drug Metabolism
• Computational Toxicology
• Machine Learning in Pharmacology

Background:

• Machine learning (ML) is increasingly applied to predict ADME (absorption, distribution, metabolism, excretion) and toxicity endpoints.
• Large datasets (>1000 compounds) are crucial for developing robust ML classification models.
• Accurate prediction of drug safety profiles is essential for efficient drug development.

Purpose of the Study:

• To review ML-driven classification studies for ADME/Toxicity endpoints from 2015-2021.
• To analyze trends in ML algorithms, dataset sizes, and validation protocols for key toxicity targets.
• To compare the performance of different ML models across various endpoints.

Main Methods:

• Comprehensive literature search and meta-analysis of classification studies.
• Focus on studies utilizing large datasets (>1000 compounds).
• Analysis of nine critical ADMET endpoints: hERG, BBB penetration, P-gp, CYP450, acute oral toxicity, mutagenicity, carcinogenicity, respiratory toxicity, and irritation/corrosion.

Main Results:

• Tree-based algorithms continue to dominate ML classification in ADMET research.
• Consensus modeling is an emerging trend for enhancing drug safety predictions.
• High-performing classification models exist for hERG cardiotoxicity and CYP450 enzyme family predictions.

Conclusions:

• ML classification models show significant progress in predicting ADMET endpoints.
• Tree-based algorithms and consensus modeling are key trends in drug safety prediction.
• Despite advancements, hERG and CYP450 remain central to ADMET research, requiring continued focus.