In the modelling of glass stability matrix iv was created by adding Tcr related properties were to matrix iii (n = 29). From each starting point (i–iv) a variable selection was performed in which input information that was not directly related to the response (i.e., noise) was removed, and thereby the predictivity and robustness of the model was increased. The accuracy of the statistically significant PLS-DA models was judged by how well the two classes of the training sets were separated from each other. In addition, for glass-forming ability, once the selection of physical properties had been finalized
the resulting models were validated with the test set. To evaluate the models for glass stability,
the fraction Selleck VX770 of the amorphous phase that had been transformed into a crystalline state upon 1 month of storage (α) was plotted against Tg, Mw, Tcr and the prediction values obtained from the PLS-DA model based on Tg and Mw. A sigmoidal relationship equation(6) α=1-1(1+e(T0-Tcr)k)was fitted to the data points in the plots by adjustment of the shape factors T0 and k. The results from the classification of glass-forming ability of the 50 compounds are presented in Table 1. For all compounds there was an agreement between DSC and X-ray data, as a clear crystallization peak visible in the thermogram upon heating in all cases coincided with a diffuse background scattering selleck inhibitor without diffraction peaks in X-ray. In the case of glibenclamide, metolazone and warfarine, the absence of both a crystallization peak and a melting peak in the DSC thermogram was taken as the sample being amorphous and stable upon heating. The X-ray analysis of these compounds confirmed they being predominantly amorphous state. Albendazole and Nifedipine showed small crystallization peaks and estimations based on the DSC-data showed that Rolziracetam were just partially amorphous (approximately 18% and 67%, respectively). Of the 50 compounds investigated, 26 were detected to be crystalline (no amorphous phase detected) after both melt-cooling and spray-drying whereas 24 showed partly or complete transformation to
the amorphous form. Hence, the latter 24 were classified as glass-formers (see Table 1). After storage for 1 month, DSC-analysis showed that 15 of the glass-formers had preserved more than 50% of its amorphous content (see Table 1). For 13 of these, the fraction crystallized was <5% which is within the uncertainty of the crystallinity determination by this method. Bicalutamide and omeprazole lost approximately 11% and 36% of their amorphous content, respectively. For the compounds classified as unstable, no amorphous phase could be detected by DSC after storage, except for griseofulvin, felodipine and acemetacin, which according to our calculations had a crystallinity of 95%, 79% and 56%, respectively, after storage.