Predicting First-in-Human Pharmacokinetics: Comparative Evaluation of Standard PBPK, High-Throughput PBPK, and...

Area of Science:

• Pharmacokinetics and Drug Development
• Computational Modeling
• Translational Science

Background:

• Accurate prediction of human pharmacokinetics (PK) is crucial for safe first-in-human (FiH) dose selection.
• Physiologically based pharmacokinetic (PBPK) modeling, high-throughput (HT) PBPK, and machine learning (ML) are key methodologies for PK prediction.
• Evaluating and comparing these methods is essential for optimizing early drug development processes.

Purpose of the Study:

• To compare the predictive accuracy of standard PBPK, HT-PBPK, and ML for human PK.
• To assess the performance of these models in predicting area under the curve (AUCinf) and maximum plasma concentration (Cmax) for oral administration.
• To explore the utility of ensemble models combining different prediction approaches.

Main Methods:

• Evaluation of 40 diverse small molecules from Roche's clinical development pipeline (2003-2024).
• Prospective assessment of standard PBPK predictions; retrospective assessment of HT-PBPK and ML predictions.
• Comparison of predicted PK parameters (AUCinf, Cmax) against observed clinical data.

Main Results:

• All three methods showed useful accuracy, with at least 49% of predictions within 2-fold of observed values.
• Standard PBPK demonstrated the highest precision, with 65% of AUCinf and 63% of Cmax predictions within 2-fold.
• HT-PBPK and ML offer viable animal-free, high-throughput alternatives, while ensemble models generally improved accuracy.

Conclusions:

• Standard PBPK remains the preferred method for FiH dose selection when highest accuracy and mechanistic insight are required.
• HT-PBPK and ML are valuable animal-free alternatives for early human dose prediction, potentially reducing costs and cycle times.
• The complementary nature of mechanistic (PBPK) and data-driven (ML) models is highlighted, suggesting combined approaches can enhance predictive power.