The use of machine learning and nonlinear statistical tools for ADME prediction

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• Drug candidate failure due to ADME properties is a significant challenge in pharmaceutical research.
• In silico prediction of ADME properties is crucial for cost reduction and efficiency enhancement.

Purpose of the Study:

• To elucidate the mechanisms and limitations of machine learning tools for ADME prediction.
• To enhance the understanding and application of machine learning in pharmaceutical research.

Main Methods:

• Development of a synthetic nonlinear dataset for 2D visualization of machine learning mechanisms.
• Application of six machine learning methods: Naive Bayes, classification and regression tree, random forest, Gaussian process, support vector machine, and k-nearest neighbor.
• Evaluation of methods across four diverse datasets.

Main Results:

• Ensemble learning and kernel machine methods demonstrated higher prediction accuracy compared to classical approaches.
• Prediction accuracy was consistent across different dataset sizes.
• Insights into machine learning mechanisms were gained through 2D visualization.

Conclusions:

• Machine learning, particularly ensemble and kernel methods, offers a powerful approach for accurate ADME prediction.
• Understanding machine learning mechanisms is vital for appropriate application in drug discovery.
• Interdisciplinary collaboration with the engineering field is beneficial for advancing these predictive tools.