Modeling and Prediction of Drug Dispersability in Polyvinylpyrrolidone-Vinyl Acetate Copolymer Using a Molecular Descriptor

Abstract

The expansion of a novel in silico model for the prediction of the dispersability of 18 model compounds with polyvinylpyrrolidone-vinyl acetate copolymer is described. The molecular descriptor R3m (atomic mass weighted 3rd-order autocorrelation index) is shown to be predictive of the formation of amorphous solid dispersions at 2 drug loadings (15% and 75% w/w) using 2 preparation methods (melt quenching and solvent evaporation using a rotary evaporator). Cosolidified samples were characterized using a suite of analytical techniques, which included differential scanning calorimetry, powder X-ray diffraction, pair distribution function analysis, polarized light microscopy, and hot stage microscopy. Logistic regression was applied, where appropriate, to model the success and failure of compound dispersability in polyvinylpyrrolidone-vinyl acetate copolymer. R3m had combined prediction accuracy greater than 90% for tested samples. The usefulness of this descriptor appears to be associated with the presence of heavy atoms in the molecular structure of the active pharmaceutical ingredient, and their location with respect to the geometric center of the molecule. Given the higher electronegativity and atomic volume of these types of atoms, it is hypothesized that they may impact the molecular mobility of the active pharmaceutical ingredient, or increase the likelihood of forming nonbonding interactions with the carrier polymer.

Interpreting the Physicochemical Meaning of a Molecular Descriptor Which Is Predictive of Amorphous Solid Dispersion Formation in Polyvinylpyrrolidone Vinyl Acetate

“…Logistic regression was used to investigate the correlations between a number of descriptors and properties and their ability to group members of the 18-molecule library according to the dispersability in PVPva. To capture all 18 APIs, these regressions were performed by combining the previously published observations of dispersability in this polymer, prepared by two different methods and at two different compositions. , As indicated in Table , univariate models constructed using the average molecular weight (AMW), the presence of halogen atoms in the API structure, the presence of hydrogen bond acceptors, and the molecular weight (MW) showed the most statistical significance for grouping molecules according to the dispersability in PVPva. Heavy atom count, log P , log S , and van der Waals volume were also significant in this respect ( p < 0.05); however, these were disregarded for their redundancy with similar descriptors (e.g., heavy atom count, which accounts for all non-H atoms in a molecule) or lack of overlap with any aspect of R3m combined with high misclassification frequency (e.g., log P , log S , and van der Waals volume).…”

“…This study focused on a library of 18 APIs (Figure ) whose dispersion potential in PVPva has been previously modeled successfully using R3m. To capture all 18 APIs, data were taken from dispersion attempts using two distinct preparation methods (melt-quenching and solvent evaporation) at two separate compositions (15% and 75% w/w with respect to the APIs) . Details of solidification, characterization, and categorization as dispersed/not dispersed in PVPva are described elsewhere. , Consistent with this previous work, the term dispersable refers to successful formation of an ASD by intimate mixing with the carrier polymer, followed by co-solidification without recrystallization .…”

“…To capture all 18 APIs, data were taken from dispersion attempts using two distinct preparation methods (melt-quenching and solvent evaporation) at two separate compositions (15% and 75% w/w with respect to the APIs) . Details of solidification, characterization, and categorization as dispersed/not dispersed in PVPva are described elsewhere. , Consistent with this previous work, the term dispersable refers to successful formation of an ASD by intimate mixing with the carrier polymer, followed by co-solidification without recrystallization . Dispersability in PVPva was modeled relative to various molecular descriptors; library molecules whose R3m > 0.65 were predicted to form dispersions in this polymer with an overall accuracy of 87.5% …”

“…To improve approaches for ASD development relative to trial and error, our group has investigated the utility of several molecular descriptors to group molecules in an 18-member API library (Figure ) according to their propensity to form an ASD in poly­(vinylpyrrolidone-vinyl acetate) (PVPva) co-polymer. , …”

Applications of Dynamic Mechanical Analysis in the Engineering of Amorphous Solid Dispersions

“…17,25,27 However, establishing a predictive mathematical relationship between mobility and stability has been challenging due to the complexity of their coupling and the lack of a fundamental understanding of key parameters controlling ASD crystallization in the solid state, 38 particularly the failure to consider supersaturation as the driving force for crystallization in the solid solution matrix of an ASD. Existing approaches are mostly empirical [39][40][41] or correlative, 42,43 considering parameters such as miscibility of drug in the polymer matrix or structure relaxation within the ASD, without much success in predicting the ASD stability . Recently, we have shown that a comprehensive model considering nucleation and growth processes can be used to describe the physical stability of ASDs in the solid state (see ►Fig.…”