Mechanistic deconvolution of <scp>BSA</scp> size variants by constrained Raman pseudo‐Voigt hard modeling during anion‐exchange chromatography

In biopharmaceutical manufacturing, protein aggregation is a critical quality attribute, necessitating rapid and reliable analytical strategies during downstream processes like anion-exchange chromatography (AEX). While Raman spectroscopy enables continuous monitoring of protein secondary structure, standard data-driven regression models struggle to decouple intrinsic structural changes from gradient-induced solvent and buffer drifts under dynamic chromatographic conditions. Addressing this methodological gap, this study establishes a constrained pseudo-Voigt hard modeling framework for the mechanistic deconvolution of bovine serum albumin (BSA) size variants during in-line Raman monitoring of AEX processes. By explicitly defining a parametric background model to capture salt-induced spectral drift, the methodology effectively isolates matrix variations from genuine protein-specific signals. The constrained hard model was applied to 285 in-line spectra across diverse chromatographic conditions, achieving reconstruction fidelity while maintaining stable, physically interpretable component identities. The mechanistically derived Amide I center of mass emerged as a robust, aggregation-sensitive descriptor that preserves structural information despite strong concentration dynamics. Furthermore, the extracted spectral features demonstrated strong predictive performance for monomer concentration and acceptable accuracy for high molecular weight components. Collectively, these results demonstrate that constrained spectral hard modeling provides a highly interpretable, robust, and calibration-light alternative to classical partial least squares approaches for the real-time monitoring of protein size variants.