Computational fluid dynamics and machine learning modeling of drug delivery by hydroxypropyl methylcellulose

This study integrates computational fluid dynamics (CFD) simulations with machine learning (ML) models to develop a framework for predicting drug release from hydroxypropyl methylcellulose (HPMC) matrices. Transient, three-dimensional CFD models were constructed to solve the conservation equations for fluid flow and mass transfer, explicitly simulating the swelling front propagation, polymer dissolution, and diffusive drug release from a cylindrical HPMC carrier into a dissolution medium. The key results from parametric CFD studies quantified the nonlinear relationship between initial drug loading and the resultant concentration gradients and release rates over time, providing a high-resolution dataset on the underlying transport phenomena. These CFD simulations were validated against experimental release data, achieving excellent agreement (mean absolute error of 1.54% and R-squared of 0.9997). The CFD-generated data, combined with experimental measurements from the literature for eight model drugs (ribavirin, ranitidine hydrochloride, isoniazid, diltiazem hydrochloride, theophylline, tinidazole, sulfamethoxazole, and propylthiouracil), were used to train the machine learning techniques. A systematic topology search identified an optimized least-squares support vector regression (LS-SVR) as the most accurate predictor. This LS-SVR model, configured with the Gaussian kernel function, successfully predicts the overall dataset with an R-squared of 0.99850, an average absolute relative deviation of 1.27%, and a root mean squared error of 0.833. This hybrid CFD-ML approach provides a powerful tool for the rational design of HPMC-based drug formulations with targeted release profiles.