Prediction of Multi-Pharmacokinetics Property in Multi-Species: Bayesian Neural Network Stacking Model with...

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Development

Background:

• Pharmacokinetic (PK) properties are critical for drug efficacy and safety.
• Predicting PK parameters across different species presents a significant challenge in drug development.

Purpose of the Study:

• To develop and validate an ensemble model for predicting PK parameters across multiple species.
• To assess the accuracy, interpretability, and application domain of the developed model.

Main Methods:

• Compiled an extensive dataset of PK parameters across various species.
• Developed the PKStack ensemble model, integrating multiple base models and incorporating uncertainty quantification.
• Collected external PK data from animal studies for validation.

Main Results:

• PKStack successfully predicted nine PK parameters across five species for 45 tasks.
• Higher prediction accuracy was observed for intravenous injections, with notable results for human Vd (R² = 0.72) and human CL (R² = 0.52).
• The model demonstrated good interpretability and defined application domains.

Conclusions:

• The PKStack model shows significant potential for practical application in drug discovery.
• Accurate prediction of PK parameters can accelerate the drug development process.
• The model's ability to handle inter-species variations and quantify uncertainty enhances its utility.