Adapting physiologically-based pharmacokinetic models for machine learning applications.

Area of Science:

• Pharmacokinetics
• Machine Learning
• Drug Development

Background:

• Physiologically-based pharmacokinetic (PBPK) models and machine learning (ML) are crucial in drug development.
• Integrating PBPK models into ML pipelines can enhance drug screening and evaluation accuracy.

Purpose of the Study:

• Develop and test a self-contained ML module to replicate summary pharmacokinetic (PK) parameters from a PBPK model.
• Evaluate the ML module's performance against a PBPK model (OpenCAT) and experimental data.

Main Methods:

• Developed an ML module to predict PK parameters using drug-specific and regimen-specific inputs.
• Utilized an open-source PBPK model, OpenCAT, for methodology demonstration.
• Tested the ML module across various drug formulations with diverse solubility and absorption characteristics.

Main Results:

• ML model predictions generally agreed within 20% of PBPK model predictions for summary PK parameters.
• Concordance was observed across a wide range of drug and formulation properties.
• Discrepancies between models and experimental data suggest potential limitations in the PBPK model itself.

Conclusions:

• The developed ML module effectively recapitulates PBPK model outputs for PK parameters.
• This integrated approach shows promise for improving predictive accuracy in drug development.
• Further refinement of PBPK models may be necessary for enhanced experimental concordance.