Мultivariate analysis for rapid screening and prediction of solid-state compatibility in pharmaceutical preformulation studies-paving the road for machine learning

Abstract

Multivariate analysis models were developed to evaluate the results obtained from a compatibility study designed for ibuprofen with a large group of different types of excipients, as a possible approach for rapid screening of the incompatibility between the active pharmaceutical ingredient (API) and excipients. The solid-state characterization of the binary mixtures was performed by Fourier transform infrared spectroscopy (FTIR) and differential scanning calorimetry (DSC). Principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) using SIMCA® software were applied for evaluation of the experimentally obtained results. The optimal PCA model for the FTIR spectra explains 96.2 % of the variations in the dataset with good statistical indicators (R2X = 0.960, Q2 = 0.900), which was also the case for the PCA model for the DSC curves (R2X = 0.981, Q2 = 0.866). The applied PLS-DA models have shown similar behaviour to the PCA. Moreover, the main spectral variations in the FTIR spectra and the thermal events in the DSC data were attributed the highest variable importance for the projection (VIP) scores in the corresponding VIP plots, confirming the model capability for predicting ibuprofen interactions. Furthermore, the prediction power of the optimal models for FTIR and DSC experimental data was evaluated by the root mean square error of prediction (RMSEP) of 0.10 and 0.16, respectively. The obtained results demonstrated the potential of multivariate statistical analysis as a machine learning-based technique for screening and prediction of ibuprofen-excipients solid-state compatibility in the preformulation phase of the pharmaceutical development of dosage forms.

Keywords

• interaction
• binary mixtures
• principal component analysis
• partial least squares-discriminant analysis
• machine learning