Human Brain Penetration Prediction Using Scaling Approach from Animal Machine Learning Models

Area of Science:

• Pharmacokinetics
• Drug Discovery
• Computational Toxicology

Background:

• Machine learning (ML) models are increasingly used for predicting drug pharmacokinetic properties.
• Previous work established ML models for predicting rat unbound brain-to-plasma ratio (Kpuu,brain).

Purpose of the Study:

• To predict human Kpuu,brain using ML models trained on animal data.
• To evaluate and compare ML models against mechanistic NeuroPK models for Kpuu,brain prediction.

Main Methods:

• Re-evaluated ML models for rat Kpuu,brain using open-source packages.
• Developed ML models for monkey Kpuu,brain prediction utilizing leave-one-out cross-validation.
• Predicted human Kpuu,brain by scaling animal ML models and compared with NeuroPK models.

Main Results:

• Rat Kpuu,brain prediction accuracy was replicated.
• The optimal ML model for monkey Kpuu,brain prediction (78% accuracy within 2-fold error) outperformed the NeuroPK model.
• ML models using rat (71% accuracy) and monkey (64% accuracy) data showed comparable predictivity to NeuroPK models for human Kpuu,brain.

Conclusions:

• ML models offer a powerful tool for predicting Kpuu,brain in animals and prospectively in humans.
• ML approaches can potentially reduce reliance on expensive primate studies for human Kpuu,brain prediction.
• These models can significantly aid drug development by providing reliable pharmacokinetic predictions early in the discovery phase.