Portal de Periódicos da CAPES

Modelling and identification of individual stage contributions in an industrial pharmaceutical process by multiblock PLS

2004; Elsevier BV; Linguagem: Inglês

10.1016/s1570-7946(04)80166-4

2543-1331

Lı́gia P. Brás, João A. Lopes, Carlos Raniery Paula dos Santos, J. P. Cardoso, José C. Menezes,

Spectroscopy and Chemometric Analyses

In this paper, multiblock partial least squares (MBPLS) was applied to model a multistage pharmaceutical process producing an active pharmaceutical ingredient (API) by fermentation. The aim is to model the final API's chromatographic purity in terms of each operating variable, establishing the relative contribution of each process stage. A balanced bootstrap algorithm was used to choose the best set of variables from each stage to obtain a model with the highest predictive capability. Seven variables from the original twenty-eight were selected and a two-latent variable MBPLS model was found to minimize the prediction residuals (QY2 = 64.3%). The multiblock methodology revealed that the fermentation stage has a prominent role in the description of the final API's chromatographic purity.