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

Amorphous solid dispersions (ASD) are an effective strategy for enhancing the solubility and bioavailability of poorly soluble drugs. However, designing and optimizing ASD formulations often rely on extensive in vitro dissolution experiments without sufficient theoretical guidance. To address this, a machine learning approach for rapidly and reliably predicting the ASD dissolution kinetics was proposed. A comprehensive data set comprising 616 dissolution profiles was collected from the "Web of Science" database, and a correlation analysis was performed to optimize input feature selection. Among the ten evaluated machine learning algorithms, lightGBM demonstrated superior predictive performance. Improvement strategies were implemented to enhance the accuracy and interpretability of the model. The improved lightGBM model achieves commendable predictive performance on commercially available ASD products, successfully quantifying the relationship between ASD formulations and the dissolution behavior. This work reduces the necessity for extensive experimental efforts and provides valuable insights into optimizing ASD formulations, thus advancing pharmaceutical formulation strategies through machine learning.