Application of Supervised and Unsupervised Learning Approaches for Mapping Storage Conditions of Biopharmaceutical Product—A Case Study of Human Serum Albumin

Abstract

Abstract The stability of biopharmaceutical therapeutics over the storage period/shelf life has been a challenging concern for manufacturers. A noble strategy for mapping best and suitable storage conditions for recombinant human serum albumin (rHSA) in laboratory mixture was optimized using chromatographic data as per principal component analysis (PCA), and similarity was defined using hierarchical cluster analysis. In contrast, separability was defined using linear discriminant analysis (LDA) models. The quantitation was performed for rHSA peak (analyte of interest) and its degraded products, i.e., dimer, trimer, agglomerates and other degradation products. The chromatographic variables were calculated using validated stability-indicating assay method. The chromatographic data mapping was done for the above-mentioned peaks over three months at different temperatures, i.e., 20°C, 5–8°C and at room temperature (25°C). The PCA had figured out the ungrouped variable, whereas supervised mapping was done using LDA. As an outcome result of LDA, about 60% of data were correctly classified with the highest sensitivity for 25°C (Aq), 25°C and 5–8°C (Aq with 5% glucose as a stabilizer), whereas the highest specificity was observed for samples stored at 5–8°C (Aq with 5% glucose as a stabilizer).