(408a) Development of an Online Free Web Application to Perform Multivariate Analysis and Determine Microcrystalline Cellulose Crystallinity from Raman Spectra
AIChE Annual Meeting
2022
2022 Annual Meeting
Particle Technology Forum
Particle Technology in Product Design and Manufacturing
Tuesday, November 15, 2022 - 3:30pm to 3:48pm
Strategies to understand and mitigate batch-to-batch raw material variability are key to ensuring process and product compliance. When critical material attributes are well-understood and quantified, operating strategies can be devised to ensure that the process will be able to cope with incoming material variability. Microcrystalline cellulose (MCC) is a raw material widely used across different industry sectors, including food and pharmaceutical. This is a semi-crystalline substance with inherent variable crystallinity (%CI) due to differences in raw material source and the different manufacturing processes through which MCC can be produced. This variance in MCC crystallinity can result in downstream process variability and should be well understood and controlled.
Aim
The aim of this study was to develop models to determine the MCC crystallinity index (%CI) from Raman spectra of 30 commercial batches using Raman probes with spot sizes of 100 mm (Kaiser Raman MR probe) and 6 mm (Kaiser Raman PhAT probe). The second aim was to pack the data and the designed models in a user-friendly online free-access tool (McCrystal â crystallinity research©) to disseminate the models to users with diverse levels of modeling experience.
Experimental
30 commercial batches of MCC were ball milled to produce amorphous samples â confirmed by Powder X-Ray Diffraction. Then, pellets of the batches as received and of the corresponding amorphous material were prepared. The pressure used to prepare these pellets was just enough to hold the powder pile together without compacting it.
Spectra of all pellets were acquired using Raman probes with spot sizes of 100 mm (Kaiser Raman MR probe, sampled on 3 different areas on one surface of the pellet) and 6 mm (Kaiser Raman PhAT probe, sampled on 1 area on the top and 1 area on the bottom of the pellet). Spectra fluorescence and baseline shifts were removed by linear-interpolation baseline and standard normal variate (SNV) normalization.
The %CI was determined using a univariate model based on the ratio of the Raman peaks at 380 and 1096 cm-1. The univariate model was adjusted for each probe. The %CI was also predicted from spectral data from each probe using partial least squares regression (PLS) models (where Raman spectra and univariate %CI were the dependent and independent variables, respectively). PLS model was initially performed using The Unscrambler X. However, all models were re-written in R, and the code written in R was compiled with the data used to build the models to be shared with the scientific community.
Results
Both univariate and PLS models showed good predictive power; the univariate model had a residual sum of squares (RSS) of 0.052 and 0.007 and Pearsonâs r of 0.895 and 0.969 for the MR and the PhAT probes, respectively, and the PLS models had a validation explained variance of 97.47 and 97.16% Bias of 0.015 and - 0.068 for the MR and the PhAT probes, respectively.
The developed PLS model substantially reduced the analysis time as it eliminates the need for spectral deconvolution.
A principal component analysis (PCA) model was used to separate the Raman spectra of the same samples captured using the different probes. This can be useful to identify non-conformity across instruments, which were encountered in this work, but also across different batches.
Conclusions
Crystallinity index was determined for 30 commercial batches of MCC using two different models, a univariate model of the ratio between two peaks and a PLS regression. A user-friendly online free-access tool, namely McCrystalâcrystallinity research©, was designed to determine crystallinity of new batches based on the models proposed in this study and Raman spectral data. McCrystal also performs MCC spectral baseline by linear interpolation, spectral normalization by SNV, and PCA. The PCA model included in this web application enables fast identification of non-conforming batches of MCC based on batch-to-batch crystallinity variation.
The web application is available at: https://sspc.ie/mccrystal-registration/