Machine learning-based unified models for predicting drug clearance from pharmacokinetic animal and study design...

Area of Science:

• Pharmacokinetics
• Machine Learning
• Computational Biology

Background:

• Drug clearance (CL) is a vital pharmacokinetic parameter for determining drug elimination rates and optimizing dosing regimens.
• Accurate CL prediction is essential for maintaining therapeutic drug concentrations and ensuring treatment efficacy.
• Existing methods for CL determination can be resource-intensive, highlighting the need for predictive modeling approaches.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting drug clearance (CL) values.
• To identify key pharmacokinetic variables that influence CL predictions.
• To assess the feasibility of using ML for cross-species extrapolation of CL values in data-scarce scenarios.

Main Methods:

• Extracted pharmacokinetic variables (drug, dose, animal species, administration route) from published literature.
• Applied nine distinct ML models to predict CL, including analysis of imbalanced and balanced datasets.
• Utilized a hybrid ML CL dataset, encompassing both true CL and CL/F values from various administration routes.

Main Results:

• Linear regression, multi-layer perceptron, and random forest models demonstrated high predictive efficiency (R2 > 0.87).
• Predictive accuracy significantly improved for specific animal groups: ungulates/small ruminants (R2 > 0.95) and companion animals (R2 > 0.92).
• The study confirmed the feasibility of predicting CL using study design variables as input parameters.

Conclusions:

• ML models offer a powerful tool for predicting drug clearance (CL) when direct data is unavailable.
• The developed models show potential for cross-species extrapolation, aiding in the estimation of CL values across different animal models.
• This research supports the use of computational and mathematical approaches to enhance pharmacokinetic predictions and inform drug disposition studies.