Extending the Use of Optical Coherence Tomography to Scattering Coatings Containing Pigments.

Coating thickness is a critical quality attribute of many coated tablets. Functional coatings ensure correct drug release kinetics or protection from light, while non-functional coatings are generally applied for cosmetic reasons. Traditionally, coating thickness is assessed indirectly via offline methods, such as weight gain or diameter growth. In the past decade, several methods, including optical coherence tomography (OCT) and Raman spectroscopy, have emerged to perform in-line measurements of various subclasses of coating formulations. However, there are some obstacles. For example, when using OCT, a major challenge is scattering pigments, such as titanium dioxide and iron oxide, which make the interface between the coating and the tablet core difficult to detect. This work explores novel OCT image evaluation techniques using unsupervised machine learning to compute image metrics. Certain image metrics of highly scattering coatings are correlated with the tablet thickness, and hence indirectly with the coating thickness. The method was demonstrated using a titanium dioxide rich coating formulation. The results are expected to be applicable to other scattering coatings and will significantly broaden the applicability of OCT to at-line and in-line coating thickness measurements of a much larger class of coating formulations.

Data-Driven Prediction of the Formation of Co-Amorphous Systems.

Co-amorphous systems (COAMS) have raised increasing interest in the pharmaceutical industry, since they combine the increased solubility and/or faster dissolution of amorphous forms with the stability of crystalline forms. However, the choice of the co-former is critical for the formation of a COAMS. While some models exist to predict the potential formation of COAMS, they often focus on a limited group of compounds. Here, four classes of combinations of an active pharmaceutical ingredient (API) with (1) another API, (2) an amino acid, (3) an organic acid, or (4) another substance were considered. A model using gradient boosting methods was developed to predict the successful formation of COAMS for all four classes. The model was tested on data not seen during training and predicted 15 out of 19 examples correctly. In addition, the model was used to screen for new COAMS in binary systems of two APIs for inhalation therapy, as diseases such as tuberculosis, asthma, and COPD usually require complex multidrug-therapy. Three of these new API-API combinations were selected for experimental testing and co-processed via milling. The experiments confirmed the predictions of the model in all three cases. This data-driven model will facilitate and expedite the screening phase for new binary COAMS.

Prediction of dissolution performance of uncoated solid oral dosage forms via optical coherence tomography.

Real-time prediction of the dissolution behavior of solid oral dosage forms is an important research topic. Although methods such as Terahertz and Raman can provide measurements that can be linked to the dissolution performance, they typically require a longer time off-line for analysis. In this paper, we present a novel strategy for analyzing uncoated compressed tablets by means of optical coherence tomography (OCT). Using OCT, which is fast and in-line capable, makes it possible to predict the dissolution behavior of tablets based on images. In our study, OCT images were obtained of individual tablets from differently produced batches. Differences between tablets or batches in these images were hardly visible to the human eye. Advanced image analysis metrics were developed to quantify the light scattering behavior captured by the OCT probe and depicted in the OCT images. Detailed investigations assured the repeatability and robustness of the measurements. A correlation between these measurements and the dissolution behavior was established. A tree-based machine learning model was used to predict the amount of dissolved active pharmaceutical ingredient (API) at certain time points for each immediate-release tablet. Our results indicate that OCT, which is a non-destructive and real-time technology, can be used for in-line monitoring of tableting processes.

A multi-step machine learning approach for accelerating QbD-based process development of protein spray drying.

This study proposes a new material-efficient multi-step machine learning (ML) approach for the development of a design space (DS) for spray drying proteins. Typically, a DS is developed by performing a design of experiments (DoE) with the spray dryer and the protein of interest, followed by deriving the DoE models via multi-variate regression. This approach was followed as a benchmark to the ML approach. The more complex the process and required accuracy of the final model is, the more experiments are necessary. However, most biologics are expensive and thus experiments should be kept to a minimum. Therefore, the suitability of using a surrogate material and ML for the development of a DS was investigated. To this end, a DoE was performed with the surrogate and the data used for training the ML approach. The ML and DoE model predictions were compared to measurements of three protein-based validation runs. The suitability of using lactose as surrogate was investigated and advantages of the proposed approach were demonstrated. Limitations were identified at protein concentrations >35 mg/ml and particle sizes of x50>6 µm. Within the investigated DS protein secondary structure was preserved, and most process settings, resulted in yields >75% and residual moisture <10 wt%.

Premexotac: Machine learning bitterants predictor for advancing pharmaceutical development.

Bitter taste receptors were recently found to be involved in numerous physiological and pathological conditions other than taste and are suggested as potential drug targets. In vivo and in vitro techniques for screening bitterants as ligands come with economical, time and ethic challenges. Therefore, in silico tools can represent a valuable alternative due to their practicality. Yet, the main challenge of already established ligand-based (LB) classifiers is the low number of experimentally confirmed bitterants and non-bitterants. Premexotac models were constructed as a LB bitterants screener, exploring novel combinations of feature extraction, feature selection and learning algorithms as a contrast with the already available screeners. Premexotac came among the top performers, exhibiting a F-1 score up to 81% on external validation. Premexotac identified as well insights on physicochemical and topological descriptors important for bitter prediction. Among the key insights, important molecular substructures from Extended Connectivity Fingerprints for bitterness classification were identified. Also, the importance of a selection of physicochemical/topological descriptors was ranked using mutual information and it was found that descriptors related to the ramification of the molecular structure and molecular weight came at the top of the ranking. The remaining challenges for improving performance were discussed and stated, widening the LB bitterness prediction outlook.