Prediction of injectables' stability leveraging inline analysis and machine learning

Aggregation and modification of the structure of biologics (i.e., proteins, mAbs, etc.) can significantly affect the quality of the final biopharmaceutical product, especially because of the increased risk of immunogenicity and the decreased specificity. The excipients in the final formulation are intended to mitigate destabilizing effects (e.g., aggregation, structural/conformational modifications, etc.). Here, we describe a novel high-throughput screening approach based on a microfluidic inline analytical platform and machine learning methods, to select the best formula composition in terms of type and concentration of excipients needed to reduce the aforementioned destabilizing effects. We show how to collect stability-related information in a fast and reliable way, with minimum material amount (i.e., few hundreds of microliters), and how to properly structure the dataset for subsequent machine learning-based processing. This approach is taught to dramatically reduce the number of experiments needed for later fine characterization of injectables' stability.