Data-Driven Design of Novel Polymer Excipients for Pharmaceutical Amorphous Solid Dispersions

Introduction: A significant branch of pharmaceutical product formulation is geared towards extending the solid-state stability of active pharmaceutical ingredients (APIs) for oral delivery by preventing API crystallization. APIs lose bioavailability and efficacy upon crystallization, which necessitates maintaining the API in an amorphous state for as long as possible until it can be administered to patients. To extend shelf stability, APIs are commonly combined with polymers as anti-plasticization agents. Polymers, along with other excipients, are formulated with APIs into amorphous solid dispersions (ASDs), which maintain APIs in a bioactive, amorphous state in pill format for oral delivery. The list of commercially used polymer excipients for ASDs is extremely limited; however, development of novel polymer excipients is hindered by the lack of quantitative structure-function relationships that define which polymer design features are desirable for extending ASD stability. Higher glass transition temperature (Tg) is associated with extended ASD stability, making Tg a potential target for ASD stability improvement. Here, we employ machine learning (ML) to study the structure-function relationships that connect polymer design with Tg by forming ASDs of polymers and probucol, our model API. Namely, we varied polymer design features such as hydrophilic feed fraction, length of monomer side chains, backbone methylation, and side chain linearity and monitored the effect of modulating these parameters on our target parameter, glass transition temperature (Tg). Experimentally derived Tg data was used to train a ML model to begin mapping structure function relationships and aid in the design of novel polymer excipients for ASDs.

Materials and

Methods: Monomers included 2-hydroxypropyl acrylate, 2-hydroxypropyl methacrylate, butyl acrylate, butyl methacrylate, tert-butyl acrylate, tert-butyl methacrylate, hexyl acrylate, cyclo-hexyl acrylate, and methyl acrylate. All monomers were polymerized via photoinduced electron/energy transfer reversible addition-fragmentation chain transfer polymerization to form polymers with degrees of polymerization fixed at 200. Polymers were film-cast with probucol at 20 wt% drug loading and Tg was determined via differential scanning calorimetry (DSC). Tg data and polymer design features were used to train an ML model with a 10-fold cross validation to predict Tg.

Results, Conclusions, and Discussions: Results and Discussions: A dataset of Tg data and polymer design features was compiled from 50 polymers tested both with and without 20 wt% probucol loading. Tg ranged from -19ºC to 107ºC depending on polymer composition and drug loading (Figure 1A, 1B). The resulting dataset contained 100 entries which were used to train the ML model to predict Tg and identify the impact of each design feature on Tg (Figure 1C). The ML model identified that the strongest predictor of Tg for the polymers in the tested library was methylation of the hydrophilic monomer. Methylation was associated with higher Tg, while hydrophilic acrylate monomers without methylation on the backbone were associated with lower Tg. The next strongest predictor was hydrophobic monomer linearity on the side chain, in which case Tg decreased when side chains were linear and increased when side chains were nonlinear (i.e., branched or cyclic). Further, bulkier side groups were predicted to increase Tg. The impact of probucol on Tg was the weakest predictive feature, with probucol predicting a slight increase on Tg.

Conclusions: This study underlines the potential for ML to streamline the materials design process and facilitate novel copolymer excipient design for ASD formulation. The ML model was able to rank design features in order of predictive power, as well as inform the effect of individual design features on Tg. We hope this work serves as a demonstration of the relative simplicity with which ML can be implemented into the materials design process. Future experiments will leverage the structure-function relationships studied in this work to develop novel polymer excipients for increased stability of ASDs.

Acknowledgements (Optional):