Answers are grounded in this paper and cite their sources.
Glycosylation is widely recognised as a critical quality attribute for therapeutic monoclonal antibodies. Yet, routine glycan analysis remains analytically challenging due to glycan heterogeneity, intensive sample preparation, prolonged analysis times, and the necessity of high-cost instrumentation. Here, we present a rapid and economical approach for quantitative estimation of N-glycans in biopharmaceutical samples by integrating attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy with machine-learning-based regression. FTIR spectra (4000−600 cm−1) of trastuzumab and bevacizumab samples, including in-process materials, were correlated with reference glycan abundances acquired using Hydrophilic Interaction Chromatography-Mass spectrometry. Sixteen clinically relevant glycans were quantitatively predicted, leveraging glycan-associated vibrational features primarily within the 1200−1000 cm−1 region. Tree-based ensemble models and convolutional neural networks, optimized via Bayesian hyperparameter tuning, attained coefficients of determination (R2) reaching 0.99 for glycans with abundances exceeding 5%, including critical glycans such as G0F-N, H4N3S1, and H4N3F1, while the majority of lower-abundance species (1−5%) were effectively captured using deep learning architectures. The proposed FTIR-machine learning framework enables rapid, non-destructive estimation of glycan composition with minimal sample preparation and is well-suited for implementation as a process analytical technology (PAT) tool for in-process monitoring of glycosylation during biopharmaceutical manufacturing.
No takes yet. Share an insight, caveat, or question.
Peruri et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: