Enabling solid oral dosage design with flexible hybrid dissolution modeling

The dissolution of oral solid dosage forms is an important quality attribute that needs to be carefully engineered to achieve the desired therapeutic effect. However, dissolution is a complex physical process that is influenced by multiple factors such as drug substance properties, formulation variables, manufacturing process conditions, and even dissolution testing conditions. Consequently, developing a robust understanding of the system is challenging and experimentally intensive. In this work, we propose a hybrid modeling framework which uses an artificial neural network (ANN) to learn a relationship between various input variables and the parameters of a mechanistic model based on the Noyes-Whitney equation. The hybrid framework offers the flexibility of data-driven approaches, while also enhancing model interpretability and capturing physical aspects of the dissolution process. The use of the framework is demonstrated via two case studies involving modified-release drug products spanning variations in formulation composition in the first and coating composition and weight gain in the second. The results establish its utility within these formulation domains. Although the framework can be deployed for other drug product types and can incorporate manufacturing routes and process conditions as model inputs, its performance in these broader settings remains to be validated. Advanced model deployment in the form of inverse design for generating formulation recommendations and transfer learning of a trained model to a new drug product system is also discussed. Overall, the proposed hybrid framework provides a practical route towards more efficient and informed formulation and process development.