Advancing pharmaceutical Intelligence via computationally Prognosticating the in-vitro parameters of fast disintegration tablets using Machine Learning models

Abstract

The field of Machine Learning (ML) has garnered significant attention, particularly in healthcare for predicting disease severity. Recently, the pharmaceutical sector has also adopted ML techniques in various stages of drug development. Tablets are the most common pharmaceutical formulations, with their efficacy influenced by the physicochemical properties of active ingredients, in-process parameters, and formulation components. In this study, we developed ML-based prediction models for disintegration time, friability, and water absorption ratio of fast disintegration tablets. The model development process included data visualization, pre-processing, splitting, ML model creation, and evaluation. We evaluated the models using root mean square error (RMSE) and R-squared score (R²). After hyperparameter tuning and cross-validation, the voting regressor model demonstrated the best performance for predicting disintegration time (RMSE: 21.99, R²: 0.76), surpassing previously reported models. The random forest regressor achieved the best results for friability prediction (RMSE: 0.142, R²: 0.7), and the K-nearest neighbor (KNN) regressor excelled in predicting the water absorption ratio (RMSE: 10.07, R²: 0.94). Notably, predicting friability and water absorption ratio using ML models is unprecedented in the literature. The developed models were deployed in a web app for easy access by anyone. These ML models can significantly enhance the tablet development phase by minimizing experimental iterations and material usage, thereby reducing costs and saving time.

Keywords: Data Engineering; Fast disintegration Tablets; K-nearest neighbor; Machine Learning; Random Forest; Voting Regressor.