Leveraging machine learning models in evaluating ADMET properties for drug discovery and development

Area of Science:

• Computational Chemistry
• Pharmacology
• Drug Discovery

Background:

• ADMET property evaluation is a major bottleneck in drug development, leading to high attrition rates.
• Traditional experimental methods for ADMET assessment are slow, expensive, and difficult to scale.
• Recent advancements in machine learning (ML) offer potential solutions to these challenges.

Purpose of the Study:

• To review the application of ML models in predicting ADMET properties.
• To explore how ML enhances accuracy and reduces experimental burden in early-stage drug development.
• To investigate the acceleration of decision-making through ML-driven ADMET prediction.

Main Methods:

• Systematic examination of ML algorithms used for ADMET prediction.
• Analysis of molecular descriptors, datasets, and model development workflows.
• Review of public databases, evaluation metrics, and regulatory aspects in computational toxicology.

Main Results:

• ML models show significant promise in predicting ADMET endpoints, often outperforming traditional QSAR models.
• These methods offer rapid, cost-effective, and reproducible alternatives for drug discovery pipelines.
• Successful case studies highlight ML applications in predicting solubility, permeability, metabolism, and toxicity.

Conclusions:

• ML is a transformative tool for early risk assessment and compound prioritization in drug discovery.
• Challenges include data quality, algorithm interpretability, and regulatory acceptance.
• Integrating ML with experimental pharmacology can significantly improve drug development efficiency and reduce failures.