A Machine Learning-Empowered Quantitative Structure-Activity Relationship Model for Predicting the Plasma Half-life...

Area of Science:

• Pharmacokinetics and Drug Development
• Computational Chemistry
• Veterinary Pharmacology

Background:

• Accurate prediction of drug plasma half-life is vital for optimizing dosage regimens in drug development.
• Pharmacokinetic data, specifically plasma half-lives, are essential for therapeutic outcome optimization.
• Existing methods for predicting drug half-life can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop machine learning-empowered Quantitative Structure-Activity Relationship (QSAR) models for predicting drug elimination half-lives in dogs.
• To utilize chemical descriptors and supervised machine learning algorithms to build predictive models.
• To establish a computational tool that supports early-stage drug development in veterinary medicine.

Main Methods:

• Collected 560 plasma half-life data points for drugs in dogs from the Food Animal Residue Avoidance Databank.
• Preprocessed pharmacokinetic data, selecting mean elimination half-life for model training.
• Employed five types of chemical descriptors and four supervised machine learning algorithms to construct QSAR models.

Main Results:

• The Deep Neural Networks model, utilizing all combined descriptor types, demonstrated the best performance.
• Achieved R-squared values of 0.80 for the fivefold cross-validation set and 0.57 for the testing set.
• Applicability domains of the trained models were visualized using Williams plots.

Conclusions:

• An effective ML-based QSAR tool was developed for predicting canine drug elimination half-lives from chemical structures.
• This predictive tool can significantly aid in the drug development process for dogs.
• The study provides a foundation for potential interspecies extrapolation of drug pharmacokinetic properties.