Physics-Informed Neural Networks for Predicting Drug Release from Biodegradable Hydrogels: A Hybrid Mechanistic-Machine Learning Approach

Biodegradable hydrogels are central to advanced drug delivery, yet the reliable prediction of drug release kinetics remains challenging due to the coupled phenomena of Fickian diffusion, polymer swelling, and hydrolytic degradation within heterogeneous matrices. Traditional mechanistic models rely on idealized assumptions, while purely data-driven machine learning approaches lack physical consistency, limiting their trustworthiness in clinical settings. This paper provides a focused review of Physics-Informed Neural Networks (PINNs) as a transformative framework that integrates the governing partial differential equations of mass transport and polymer degradation directly into the loss function of deep neural networks. The architecture enforces conservation laws and boundary conditions, enabling accurate release predictions even from sparse experimental data. We outline the polymer physics of hydrogels, formulate the PINN for spherical degradable systems, and discuss inverse design, transfer learning, and digital twin applications for personalized dosing. Through illustrative case studies on PLGA and chitosan gels, we demonstrate that PINNs outperform conventional neural networks in accuracy, physical rigor, and data efficiency. Key challenges – lack of standardized polymer databases, interpretability for regulatory acceptance, and extension to multi-stimuli systems are highlighted. The review underscores the potential of PINNs to bridge polymer engineering, pharmaceutical sciences, and clinical translation.

کلیدواژه ها:

drug release ، biodegradable hydrogels ، Physics-Informed Neural Network ، polymer degradation ، machine learning ، inverse design ، digital twin