Data-driven insights into the properties of liquisolid systems based on machine learning algorithms

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Materials Science

Background:

• Liquisolid systems (LS) enhance drug dissolution and bioavailability by converting liquid drugs into powders.
• Evaluating factors influencing LS performance is crucial for formulation development.
• Existing research highlights the need for comprehensive analysis of LS characteristics.

Purpose of the Study:

• To assess the applicability of machine learning (ML) algorithms for evaluating liquisolid systems.
• To identify critical factors governing liquisolid system performance using literature data.
• To develop predictive models for liquisolid system characteristics.

Main Methods:

• A dataset of 425 liquisolid formulations was compiled from published literature.
• Machine learning algorithms (Gradient Boosting, Adaptive Boosting, Random Forest) were employed.
• Analysis focused on preparation methods, formulation parameters, and system properties like flowability, compact hardness, and drug dissolution.

Main Results:

• Novel preparation methods (fluid bed, extrusion/spheronization) and excipients improved LS properties.
• Key formulation factors identified include carrier/coating agent type and content, liquid load, drug type and content, and preparation method.
• Developed ML models demonstrated high prediction accuracy (>80%) on test data.

Conclusions:

• Machine learning models offer valuable insights into critical attributes affecting liquisolid system performance.
• ML can serve as a powerful tool for the development and optimization of liquisolid systems.
• This study validates the use of ML for data-driven evaluation in pharmaceutical formulation.