Predictive modeling of controlled drug release from polysaccharide-based systems using gradient boosting and...

Area of Science:

• Pharmaceutical Science
• Computational Chemistry
• Materials Science

Background:

• Accurate prediction of drug release kinetics is crucial for designing effective polysaccharide-based drug delivery systems.
• Existing methods often lack the precision needed for rational formulation development.
• Understanding the relationship between polymer structure and drug release is key.

Purpose of the Study:

• To develop a hybrid machine learning framework integrating Raman spectroscopy and formulation descriptors to model drug release profiles.
• To optimize and evaluate the predictive performance of advanced machine learning models for drug release kinetics.
• To identify key spectral features and formulation parameters influencing drug release.

Main Methods:

• A dataset of 155 experimental instances across 13 formulation groups was utilized, including Raman spectral data, medium descriptors, and temporal information.
• Feature selection identified 17 informative Raman peaks, combined with medium and time as model inputs.
• Four hybrid predictors (XGSO, XGQO, LGSO, LGQO) were developed by optimizing Extreme Gradient Boosting (XGB) and Light Gradient Boosting (LGB) models using Swarm-Assisted Bayesian Optimization (SABO) and Quantum-Inspired Optimization (QIO).

Main Results:

• The optimized hybrid models significantly outperformed single learners in predicting drug release kinetics.
• The XGSO and LGSO models achieved the lowest prediction errors, with test RMSE values of 0.065 and 0.077, and R² values of 0.961 and 0.939, respectively.
• SHapley Additive exPlanations (SHAP) highlighted the strong influence of Raman bands (940-990 cm⁻¹ and 470-510 cm⁻¹) related to glycosidic backbone vibrations, along with time and medium effects.

Conclusions:

• The proposed hybrid machine learning framework accurately predicts drug release kinetics from polysaccharide matrices.
• The study successfully linked specific polysaccharide structural vibrations (via Raman spectroscopy) to macroscopic drug release behavior.
• This data-driven approach offers a powerful tool for optimizing drug delivery formulations and advancing rational design principles.