Low-dimensional neural ODEs and their application in pharmacokinetics.

Area of Science:

• Pharmacology and Pharmacometrics
• Computational Biology
• Machine Learning Applications

Background:

• Machine Learning (ML) is increasingly integrated into scientific research.
• Neural Ordinary Differential Equations (NODEs) represent a novel ML tool for dynamical systems.
• NODEs offer a new approach to pharmacokinetic (PK) and pharmacodynamic (PD) modeling.

Purpose of the Study:

• Introduce the functionality of NODEs for PK/PD modeling.
• Develop low-dimensional NODE structures based on PK principles.
• Address challenges like overfitting and extrapolation in NODE application.

Main Methods:

• Developed specific low-dimensional NODE structures informed by PK principles.
• Applied NODEs to various PK modeling scenarios (multi-compartmental, TDD, delayed absorption).
• Investigated and proposed solutions for NODE challenges: overfitting and extrapolation.

Main Results:

• NODEs effectively described PK data across diverse scenarios.
• Proposed NODE structures demonstrated ability to simulate data for new subjects.
• Successfully illustrated NODE application in PK analyses and integration with mechanistic models.

Conclusions:

• Low-dimensional NODEs provide a viable and effective tool for PK analyses.
• NODEs show significant potential for advancing pharmacology and pharmacometrics.
• This work enhances understanding and practical application of NODEs in drug modeling.