Deep-PK: deep learning for small molecule pharmacokinetic and toxicity prediction

Area of Science:

• Computational chemistry
• Drug discovery
• Pharmacokinetics and toxicity prediction

Background:

• Drug development requires evaluating pharmacokinetic properties (absorption, distribution, metabolism, excretion, and toxicity - ADMET).
• Traditional in vitro, in vivo, and pre-clinical ADMET data acquisition is costly and time-consuming.
• Existing computational methods often lack accuracy, interpretability, and broad applicability for diverse targets.

Purpose of the Study:

• To introduce Deep-PK, a novel deep learning platform for predicting, analyzing, and optimizing pharmacokinetic and toxicity properties.
• To address limitations in current computational approaches for ADMET prediction.

Main Methods:

• Application of graph neural networks and graph-based signatures as graph-level features.
• Development of a deep learning model trained on 73 diverse endpoints (64 ADMET and 9 general properties).

Main Results:

• Achieved high predictive performance across a wide range of pharmacokinetic and toxicity endpoints.
• Demonstrated the platform's capability in supporting molecular optimization and interpretation.

Conclusions:

• Deep-PK offers a powerful, accurate, and user-friendly solution for pharmacokinetic and toxicity prediction in drug development.
• The platform aids researchers in optimizing and understanding molecular properties, accelerating the drug discovery pipeline.