فهرست نشریات علمی دسترسی آزاد ایران

This study investigates the application of Deep Neural Networks (DNNs) to optimize the formulation, development, and performance evaluation of Metformin Hydrochloride sustained-release tablets, a key medication for managing Type 2 diabetes. Traditional drug formulation methods are often time-consuming and constrained by the complexity of the formulation process. This research addresses these challenges by utilizing DNNs to create predictive models that accurately forecast critical formulation outcomes, such as dissolution rates. The study began with a comprehensive pre-formulation analysis to assess the physicochemical properties of Metformin Hydrochloride and its compatibility with various excipients. Using a 3² factorial Design of Experiments (DoE) approach, 24 formulations (F1–F24) were prepared through the wet granulation method, varying the concentrations of Polyox WSR 303 and Povidone K30. The tablets were evaluated for post-compression parameters and in vitro dissolution performance. Experimental data from these formulations were used to train a DNN model to predict optimal formulation parameters based on performance metrics. Among the formulations, the DNN identified F1 as the optimal formulation, predicting a drug release of 99.23%. Experimental validation of F1 revealed an in vitro drug release of 98.95%, closely matching the predicted value. The optimal composition included 95 mg of Polyox WSR 303 and 115 mg of Povidone K30. A comparison with a computerized simulation model showed a difference factor (f1) of 1.71 and a similarity factor (f2) of 91.48, confirming a high degree of similarity between the dissolution profiles. This study highlights the potential of deep learning to streamline pharmaceutical development, improve formulation precision.