Toward Generalizable Data-Driven Pharmacokinetics with Interpretable Neural ODEs.

Area of Science:

• Pharmacokinetics and Computational Biology
• Drug Development and Precision Medicine
• Artificial Intelligence in Healthcare

Background:

• Accurate drug concentration-time (C-t) profile modeling is crucial for drug development and personalized dosing.
• Traditional pharmacokinetic (PK) models face limitations in scalability and adaptability due to rigid assumptions and extensive parametrization.
• There is a need for advanced modeling approaches that can handle diverse compounds and patient populations effectively.

Purpose of the Study:

• To introduce Uni-PK, a unified neural framework for end-to-end pharmacokinetic modeling.
• To develop a scalable and interpretable solution for predicting drug concentration dynamics.
• To enable personalized preclinical and clinical applications by incorporating interindividual variability.

Main Methods:

• Developed Uni-PK by combining molecular representations with neural ordinary differential equations (NODEs) within a PK structure.
• Employed a flexible context encoder to integrate auxiliary covariates (e.g., species, dosing regimen) for personalized modeling.
• Enabled direct dynamic trajectory modeling of drug concentrations from molecular and individual inputs, facilitating learning in data-scarce conditions.

Main Results:

• Uni-PK demonstrated robust performance on rat and human datasets across various administration routes and physiological states.
• The framework showed consistency with established pharmacokinetic principles, validating its mechanistic grounding.
• Achieved end-to-end learning capabilities, even under data-scarce and noisy conditions, outperforming traditional methods.

Conclusions:

• Uni-PK offers a scalable, interpretable, and animal-sparing solution for next-generation pharmacokinetic modeling.
• The integration of chemical structure and individual-specific information advances precision therapeutics.
• This unified neural framework has the potential to significantly impact drug development and individualized dosing strategies.