Machine Learning Prediction and Validation of Plasma Concentration-Time Profiles.

Area of Science:

• Pharmacology
• Computational Biology
• Data Science

Background:

• Machine learning (ML) is increasingly used for covariate selection in population pharmacokinetics (PPK).
• Limited research exists on predicting drug plasma concentration profiles using nonlinear mixed-effect models combined with ML.
• Existing studies often lack validation for prediction accuracy across diverse patient populations and dosing conditions.

Purpose of the Study:

• To address the gap in predicting drug plasma concentration profiles using ML and nonlinear mixed-effect models.
• To validate the prediction accuracy and applicability of ML models with virtual and real-world data.
• To evaluate the potential of ML as a complementary tool in pharmacokinetic and pharmacodynamic (PK/PD) studies.

Main Methods:

• Utilized remifentanil as a model drug for predicting plasma concentration profiles.
• Applied machine learning models, specifically Random Forest, for prediction tasks.
• Generated virtual training datasets through clustering based on test dataset size and diversity.
• Validated models using both simulated (virtual) and actual patient (real-world) data.

Main Results:

• Random Forest models demonstrated high prediction accuracy for both virtual and real-world datasets.
• ML models proved effective for large-scale datasets with variable dosing times and amounts.
• The approach showed applicability across diverse patient populations and dosing conditions.

Conclusions:

• Machine learning models offer a highly accurate and efficient method for predicting drug plasma concentration profiles.
• ML provides a valuable, fit-for-purpose approach that complements traditional PPK methods.
• This study highlights the potential of ML to advance future pharmacokinetic and pharmacodynamic research.