Predicting curcumin release kinetics from nanocarriers using a physics-informed machine learning framework.

Area of Science:

• Nanomedicine
• Materials Science
• Computational Chemistry

Background:

• Curcumin nanoformulations exhibit variable release kinetics, hindering predictable drug delivery.
• Existing models lack a unified framework to analyze diffusion across diverse nanocarrier materials.
• Limited datasets and inconsistent experimental conditions impede understanding of curcumin release parameters.

Purpose of the Study:

• To develop a physics-informed machine learning framework for predicting curcumin release kinetics.
• To achieve high physical consistency in predictions across heterogeneous nanocarriers.
• To establish a foundation for data-driven engineering of curcumin delivery systems.

Main Methods:

• A dataset of 75 curcumin nanoformulations was analyzed using 13 physicochemical descriptors.
• Release kinetics were modeled using a two-component diffusion model.
• A Physics-Informed Neural Network (PINN) with monotonicity and non-negativity constraints was trained and validated.

Main Results:

• The PINN model achieved 0.92 R² and 98% physical consistency, outperforming baseline models.
• External validation with 31 independent formulations yielded 0.885 R².
• Design-space mapping identified optimal parameters for rapid, balanced, and sustained release profiles.

Conclusions:

• The developed framework accurately predicts kinetic parameters for diverse nanocarriers.
• This approach facilitates data-driven formulation engineering for curcumin delivery systems.
• The methodology paves the way for automated design of controlled-release nanomedicines.