Advancing Amorphous Solid Dispersions Design: Insights into Dissolution Kinetics via Thermodynamic Descriptor and...

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Materials Science

Background:

• Amorphous solid dispersions (ASDs) improve solubility and bioavailability of poorly soluble drugs.
• Current ASD formulation optimization relies heavily on time-consuming in vitro dissolution testing.
• Lack of theoretical guidance hinders efficient ASD design.

Purpose of the Study:

• To develop a machine learning model for rapid and reliable prediction of ASD dissolution kinetics.
• To reduce the need for extensive experimental screening in ASD formulation.
• To provide theoretical insights into ASD formulation-dissolution behavior relationships.

Main Methods:

• Collected a dataset of 616 dissolution profiles from scientific literature.
• Performed correlation analysis for optimal input feature selection.
• Evaluated ten machine learning algorithms, with lightGBM showing the best performance.
• Implemented model improvement strategies for accuracy and interpretability.

Main Results:

• LightGBM demonstrated superior predictive performance for ASD dissolution kinetics.
• The optimized model accurately predicted dissolution behavior for commercial ASD products.
• Quantified the relationship between ASD formulation characteristics and dissolution.

Conclusions:

• Machine learning, specifically lightGBM, offers a powerful tool for predicting ASD dissolution.
• This approach significantly reduces experimental workload in pharmaceutical formulation.
• Provides valuable insights for advancing the design and optimization of ASDs.