Data-Smart Machine Learning Methods for Predicting Youngs Modulus of Directly Compressed Blends of Pharmaceutical Powders

The ability to predict the mechanical properties of compacted powder blends of Active Pharmaceutical Ingredients (API) and excipients solely from those of the individual components can reduce the amount of “trial-and-error” involved in formulation design. Machine Learning (ML) models can reduce model development time and effort with the imperative of adequate historical data. This work evaluates linear and non-linear ML for predicting the Youngs Modulus of directly compressed arbitrary blends of known excipients and API from readily available information about the API. The training data obtained from three BCS Class I APIs and four excipients demonstrate “data-smart” strategies to train ML models efficiently. The results indicate that even simple linear ML model provide reasonable prediction accuracy. The practical benefits of this method and how it compares with other mechanistic models are discussed. Finally, we demonstrate an application of the model to enable Quality-by-Design in drug product development.