Monitoring an Aromatic Amide Active Pharmaceutical Ingredient in a Reactive Crystallization by in situ Raman Spectroscopy: Comparing Data Analysis between Partial Least Squares Regression and Non-Negative Matrix Factorization

Abstract Continuous crystallizations of advanced pharmaceutical ingredients and their intermediates are gaining attention within the pharmaceutical industry. Conventional batch crystallization processes can be converted to continuous flow for process intensification. Selecting the appropriate process analytical tools is key to process optimization. In this work, we present an approach to monitor the reactive crystallization of an aromatic amide (AA) in situ using a Raman probe. One of the key challenges was to distinguish between the AA in the solid form, a neutral solute, and the anion salt, all of which have very similar structures. Two multivariate analyses were compared, namely a partial least squares regression (PLSR) and non-negative matrix factorization (NMF). The PLSR predicted concentration of the total AA and its limitations in separating each of the three forms of AA due to the effects of supersaturation, nucleation, and crystallization at the crossover point of the reactive crystallization are discussed. We subsequently apply the NMF to the Raman spectra. The NMF is an unsupervised method, uses non-negativity values to match the physical reality of the Raman spectra, and has, for the first time, provided the separated spectrum for each of the three forms: the dissolved anion AAanion salt, the neutral dissolved solute AAsolute, and the solid form AAcrys, adding insights into the molecular changes occurring throughout the reactive crystallization and allowing continuous process monitoring and control.