Development of prediction model with machine learning in continuous twin screw granulation

Purpose This study aimed to use machine learning to predict the quality of a drug product prepared through continuous manufacturing. Methods A design of experiment (DoE) was employed to identify the relationship between process parameters in continuous twin screw granulation and intermediate and drug product quality. Subsequently, unsupervised and supervised machine learning were applied to develop a dissolution prediction model. Hierarchical clustering analysis (HCA) was conducted to evaluate the similarity between the datasets generated by the DoE. In addition, supervised learning, such as partial least squares (PLS) regression with variable importance in projection (VIP), was conducted, and a prediction model was developed using an artificial neural network (ANN) as a machine learning model. Moreover, principal component analysis (PCA), which is an unsupervised learning method, was used to preprocess the ANN model to reduce the number of features. Results HCA divided the data set obtained by DoE into three clusters, and it was confirmed that the L/S ratio significantly affected granule and tablet properties. The ANN, PLS-VIP(0.8)-ANN, PLS-VIP(1.0)-ANN, and PCA-ANN models were developed and statistically compared. The PLS-VIP(0.8)-ANN accurately predicted the dissolution profile with a high coefficient of determination and low mean absolute deviation, root mean square error, and sum of squared errors. Conclusion This paper describes the application of various machine learning algorithms to predict drug product quality as a control strategy of continuous manufacturing.