Hierarchical deep compartment modeling: A workflow to leverage machine learning and Bayesian inference for...

Area of Science:

• Pharmacometrics
• Machine Learning in Drug Development
• Computational Pharmacology

Background:

• Population pharmacokinetic (PK) modeling is crucial for understanding drug disposition in diverse patient groups.
• Covariate modeling in PK analysis is complex, with challenging relationships between covariates and PK parameters.
• Previous machine learning approaches like deep compartment modeling (DCM) had limitations in assessing errors, variability, and uncertainty.

Purpose of the Study:

• To introduce Hierarchical Deep Compartment Modeling (HDCM) as an advancement over DCM for PK analysis.
• To leverage machine learning for improved covariate and PK parameter relationship elucidation.
• To enable simultaneous evaluation of random effects and quantification of model uncertainty.

Main Methods:

• Application of HDCM, a machine learning-driven approach.
• Integration of Bayesian inference for uncertainty quantification.
• Utilizing open-source Julia tools for workflow implementation.

Main Results:

• HDCM effectively discerns complex interplay between covariates and PK parameters.
• Simultaneous evaluation of random effects and uncertainty quantification are achieved.
• The HDCM workflow provides a robust framework for advanced PK modeling.

Conclusions:

• HDCM overcomes limitations of DCM, offering enhanced predictive accuracy and model reliability.
• This method provides a powerful tool for understanding drug behavior and variability in populations.
• The tutorial facilitates practical application of HDCM in pharmacokinetic research.