Adaptive AI framework for pharmacokinetics using GATs, transformers, and AutoML.

Area of Science:

• Pharmacokinetics
• Artificial Intelligence in Drug Discovery
• Computational Chemistry

Background:

• Accurate prediction of pharmacokinetic parameters (absorption, distribution, metabolism, excretion) is crucial but challenging in drug discovery.
• Traditional experimental methods are time-consuming and expensive, hindering rapid decision-making.
• Existing computational models are often static and require extensive retraining with new data.

Purpose of the Study:

• To develop and validate a real-time artificial intelligence (AI) framework for predicting pharmacokinetic parameters.
• To enable dynamic model recalibration using newly available data without full retraining.
• To enhance the scalability, responsiveness, and predictive accuracy of pharmacokinetic modeling in drug development.

Main Methods:

• Integration of graph attention networks, Transformer models, and automated machine learning.
• Development of a dynamic framework that periodically incorporates new data, stratified by administration routes (e.g., intravenous, oral).
• Implementation of model recalibration without complete retraining to adapt to new compound data, including single-point measurements.

Main Results:

• Achieved a mean coefficient of determination (R²) of 0.93 for pharmacokinetic parameter predictions.
• Attained a mean absolute error (MAE) of 0.059, indicating high predictive accuracy.
• Demonstrated superior performance compared to conventional batch learning techniques in dynamic prediction scenarios.

Conclusions:

• The proposed AI framework offers a promising solution for real-time, accurate pharmacokinetic parameter prediction.
• The dynamic recalibration approach enhances flexibility and allows integration of new compounds efficiently.
• This framework has the potential to significantly accelerate data-driven decision-making throughout the drug development pipeline.