Modernizing non-classical protein crystallization through industry 4.0: Advanced monitoring and modelling utilizing process analytical technology

Abstract

Modernizing manufacturing processes of pharmaceutical drug products with advanced monitoring and modelling can aid in the transition towards Industry 4.0 with the benefit of increased productivity. This study investigated the use of process analytical technology in combination with partial least squares (PLS) regressions to create two soft sensors to predict the mass fraction of crystallised active pharmaceutical ingredient (API) and mass fraction of dissolved API during a non-classical protein crystallization with amorphous precursors. The PLS model for predicting the amount of crystalline API was based on Raman spectra, chord length distributions and turbidity data using small-angle X-ray scattering as a reference method. The model had a root mean square error on cross-validation (RMSECV) of 5 %. The model predicting mass fraction of dissolved API was based only on the Raman spectra and used high performance liquid chromatography as reference method. This model had a RMSECV of 3 % A two-step nucleation model was fitted to the predictions from the sensors and showed good agreement between data and model with a root mean square error of 2 %.