Key points are not available for this paper at this time.
Abstract Direct compression (DC) remains a popular manufacturing technology for producing solid dosage forms. However, the formulation optimisation is a laborious process, costly and time-consuming. The aim of this study was to determine whether machine learning (ML) can be used to accelerate developments by predicting the drug release profiles under dynamic conditions given the composition of formulations. A total of 377 formulations were produced in-house and their release profile under dynamic dissolution conditions was measured from 0 to 480 min across 11 time points. A subsequent ML analysis involved predicting the entire release profile. Six different ML techniques were explored, where random forest (RF) and extreme gradient boosting (XGB) were found to achieve a fivefold cross-validation R 2 of 0.635 ± 0.047 and 0.601 ± 0.091, respectively. A second ML strategy was developed, where the ML techniques predict the kinetic parameters of the Weibull and a modified first-order kinetic release model and subsequently use the predicted parameters to fit the release profiles. The R 2 results using RF were comparable to the first strategy. These findings demonstrate that ML can be used to predict entire drug release profiles during dynamic dissolution studies, whilst simultaneously providing insight into kinetic parameters, thus making the modelling process more informative for pharmaceutical researchers. Future work will seek to investigate more ‘kinetic-informed’ ML models. Graphical Abstract
Protopapa et al. (Mon,) studied this question.