Prediction drug release profile from chitosan nanoparticles: integration of experimental data and machine learning models.

Abstract

Background: Integration of artificial intelligence such as machine learning into pharmaceutical sciences has become increasingly necessary since rapid progress of novel drug delivery systems and increasing demand to accelerate research while reducing time and cost. According to the controllable nature of nanoparticle characteristics, ML algorithms offer a promising approach to predict critical properties of drug delivery systems including drug release profiles.

Objective: This study aimed to develop and investigate the machine learning models for prediction of cumulative drug release profile from chitosan nanoparticles, based on different physicochemical parameters in formulation.

Methods: In this study, we extracted experimental data from 115 research articles published between 2000 and 2020 with focus on chitosan nanoparticles prepared by ionic gelation method. For prediction of cumulative drug release profiles at multiple time points, after curating 190 datapoints with appropriate physicochemical parameters, we developed and evaluated two supervised ML models including Random Forest Regression and XGBoost using R² and MSE as evaluation parameters. Additionally, to improve model performance, we used feature importance analysis to identify and remove less influential variables.

Results: The results demonstrated that Random Forest Regression consistently outperformed XGBoost at the most time points. Furthermore, some variables like release medium temperature and drug solubility were excluded in a refined models since minimum contribution in model accuracy. The refined models showed improvement in prediction performance.

Conclusion: These findings highlight the value of ML-based modeling in pharmaceutical formulation and emphasis its potential as a powerful tool for design and optimization of nanobased drug delivery systems.