拉曼光谱在线监测蛋白A亲和层析过程中抗体浓度变化

打开文本图片集

中图分类号：TQ028.8 文献标志码：A DOI：10.3969/j.issn.1003-9015.2025.02.009

文章编号：1003-9015（2025）02-0263-10

Abstract：To determine the antibody concentration incellculture broth during protein A afinitychromatography， Raman spectrum was used anda model was established for on-line measurement of antibody concentration in complex feedstock．The partial least squares （PLS）regresson and ridge regression models were investigated with the Raman spectral data under diferent loading conditions.The resultsindicate thatthe PLS model had a minimum Root Mean Square Error of Cross-Validation （RMSECV）of O.088 g·L-1anda Root Mean Square Error of Prediction（RMSEP）of 0.112g⋅L-1 . In comparison， the ridge regresson model achieved an RMSECV of 0.097 g⋅L-1 （204 and an RMSEP of 0.055g⋅L-1 . The ridge regression model could balance the model complexity and accuracy，reduce the risk of overfiting and improve predictive performance.Additionally，with various data splits，the ridge regression was always beter than PLS in both fiting and prediction. The results demonstrate thatthe ridge regressionbased Raman spectrum model can accurately predict antibody concentration in complex fedstock during the protein A afinity chromatography，ofering a promising tol foronline monitoring in continuous antibody capture process.

Key words：Raman spectroscopy； affinity chromatography；online monitor；partial least squares；ridge regression

1前言

近年来，抗体药物发展迅速，产能不断扩大，但是传统的批次生产存在固有局限，投资巨大、风险高、设备效率低，难以适应未来生产的需要。（剩余13064字）