Integration of wavelet scattering transform and artificial neural networks for online monitoring of antibody aggregates by Raman spectrum

doi:10.11949/0438-1157.20250612

Abstract

Abstract:

Online monitoring the product quality attributes is critical in biopharmaceutical manufacturing. Although Raman spectrum has emerged as a promising online monitoring tool, the inherent complexity of its spectral signals poses significant challenges for the prediction of the product quality attributes. This study developed a novel modeling method that integrates wavelet scattering transform and artificial neural network (ST-ANN) to dynamically monitor the concentration of antibody monomer and aggregate for flow-through chromatography. Compared with conventional approaches, such as partial least squares regression, K-nearest neighbors, artificial neural networks, and convolutional neural networks, the ST-ANN model showed an improved predictive performance with the determination coefficients of 0.982 for both monomer and aggregate in test sets. The ST-ANN model had an improved predictive capacity, which is promising as new modeling method for online monitoring with Raman spectrum in biopharmaceutical downstream bioprocesses.

Key words: Raman spectrum, online monitoring, artificial neural networks, wavelet scattering transform, monoclonal antibody, antibody aggregate

摘要：

在生物制药过程中，实时监测产品质量属性至关重要，拉曼光谱作为一种在线监测技术，面临复杂信号带来的建模挑战。本文提出了一种结合小波散射变换和人工神经网络（ST-ANN）的建模方法，用于在线监测流穿模式层析过程中的抗体单体与聚集体浓度变化。结果表明，相较于传统的偏最小二乘回归、K-最近邻、人工神经网络和卷积神经网络模型，ST-ANN对验证集的预测误差显著降低，单体和聚集体的测试集决定系数R²均达到0.982。ST-ANN展现出更强的泛化能力，为拉曼光谱应用于生物制药下游分离过程在线监测提供了新的建模方法。

关键词: 拉曼光谱, 在线监测, 人工神经网络, 小波散射变换, 单克隆抗体, 抗体聚集体

CLC Number:

TQ 028.8

Akang TONG, Liangzhi QIAO, Dong GAO, Haibin WANG, Haibin QU, Shanjing YAO, Dongqiang LIN. Integration of wavelet scattering transform and artificial neural networks for online monitoring of antibody aggregates by Raman spectrum[J]. CIESC Journal, DOI: 10.11949/0438-1157.20250612.

童阿康, 乔亮智, 高栋, 王海彬, 瞿海斌, 姚善泾, 林东强. 结合小波散射变换和人工神经网络的拉曼光谱在线监测抗体聚集体[J]. 化工学报, DOI: 10.11949/0438-1157.20250612.

Figures/Tables 9

Fig.1 Scheme of wavelet scattering transform

Table 1 Concentrations of mAb monomers and aggregates in the chromatographic experiments

实验序号	单抗总浓度 (g/L)	聚集体含量 (%)
1	12.48	7.20
2	11.19	7.40
3	10.41	8.45
4	10.13	8.87
5	8.18	10.56

Fig. 2 Comparison of Raman spectral data before and after preprocessing

Fig. 3 Comparison of wavelet scattering transform for Raman spectrum

Fig. 4 Modeling performance of training and validation sets for mAb monomers with different models

Fig. 5 Modeling performance of training and validation sets for mAb aggregates with different models

Table 2 Comparison of real-time Raman spectrum monitoring in mAb chromatographic purification process

模型	研究对象	光谱采集时间 s	MAE g/L	文献
KNN	单体聚集体	22.5	1.08 1.05	[9]
CNN	聚集体	38	0.76	[10]
2D-KNN	单体聚集体	30	3.40 0.10	[31]
ST-ANN	单体聚集体	12	0.34 0.03	本研究

Fig.6 Prediction performance of CNN and ST-ANN models for mAb monomer and aggregate concentrations

Fig.7 Comparison of the predicted breakthrough curves with CNN and ST-ANN models to the experimental data

References 32

[1]	Chan A C, Carter P J. Therapeutic antibodies for autoimmunity and inflammation[J]. Nature Reviews Immunology, 2010, 10(5): 301-316.
[2]	Vázquez-Rey M, Lang D A. Aggregates in monoclonal antibody manufacturing processes[J]. Biotechnology and Bioengineering, 2011, 108(7): 1494-1508.
[3]	Bansal R, Gupta S, Rathore A S. Analytical platform for monitoring aggregation of monoclonal antibody therapeutics[J]. Pharmaceutical Research, 2019, 36(11): 152.
[4]	Arosio P, Barolo G, Müller-Späth T, et al. Aggregation stability of a monoclonal antibody during downstream processing[J]. Pharmaceutical Research, 2011, 28(8): 1884-1894.
[5]	São Pedro M N, Klijn M E, Eppink M H, et al. Process analytical technique (PAT) miniaturization for monoclonal antibody aggregate detection in continuous downstream processing[J]. Journal of Chemical Technology & Biotechnology, 2022, 97(9): 2347-2364.
[6]	Makki A A, Massot V, Byrne H J, et al. Understanding the discrimination and quantification of monoclonal antibodies preparations using Raman spectroscopy[J]. Journal of Pharmaceutical and Biomedical Analysis, 2021, 194: 113734.
[7]	Guo S X, Beleites C, Neugebauer U, et al. Comparability of Raman spectroscopic configurations: a large scale cross-laboratory study[J]. Analytical Chemistry, 2020, 92(24): 15745-15756.
[8]	Yilmaz D, Mehdizadeh H, Navarro D, et al. Application of Raman spectroscopy in monoclonal antibody producing continuous systems for downstream process intensification[J]. Biotechnology Progress, 2020, 36(3): e2947.
[9]	Wang J R, Chen J Y, Studts J, et al. Automated calibration and in-line measurement of product quality during therapeutic monoclonal antibody purification using Raman spectroscopy[J]. Biotechnology and Bioengineering, 2023, 120(11): 3288-3298.
[10]	Wang J R, Chen J Y, Studts J, et al. In-line product quality monitoring during biopharmaceutical manufacturing using computational Raman spectroscopy[J]. mAbs, 2023, 15: 2220149.
[11]	Shen J H, Li M, Li Z F, et al. Single convolutional neural network model for multiple preprocessing of Raman spectra[J]. Vibrational Spectroscopy, 2022, 121: 103391.
[12]	Ho C S, Jean N, Hogan C A, et al. Rapid identification of pathogenic bacteria using Raman spectroscopy and deep learning[J]. Nature Communications, 2019, 10: 4927.
[13]	Poth M, Magill G, Filgertshofer A, et al. Extensive evaluation of machine learning models and data preprocessings for Raman modeling in bioprocessing[J]. Journal of Raman Spectroscopy, 2022, 53(9): 1580-1591.
[14]	Nitika N, Keerthiveena B, Thakur G, et al. Convolutional neural networks guided Raman spectroscopy as a process analytical technology (PAT) tool for monitoring and simultaneous prediction of monoclonal antibody charge variants[J]. Pharmaceutical Research, 2024, 41(3): 463-479.
[15]	Liu Y, Brown S D. Wavelet multiscale regression from the perspective of data fusion: new conceptual approaches[J]. Analytical and Bioanalytical Chemistry, 2004, 380(3): 445-452.
[16]	Rolinger L, Rüdt M, Hubbuch J. Comparison of UV- and Raman-based monitoring of the Protein A load phase and evaluation of data fusion by PLS models and CNNs[J]. Biotechnology and Bioengineering, 2021, 118(11): 4255-4268.
[17]	Mallat S. Group invariant scattering[J]. Communications on Pure and Applied Mathematics, 2012, 65(10): 1331-1398.
[18]	Lone A W, Aydin N. Wavelet scattering transform based Doppler signal classification[J]. Computers in Biology and Medicine, 2023, 167: 107611.
[19]	Bruna J, Mallat S. Invariant scattering convolution networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(8): 1872-1886.
[20]	Ding Z W, Zhang C F, Huang X, et al. Recognition method of coal–rock reflection spectrum using wavelet scattering transform and bidirectional long–short-term memory[J]. Rock Mechanics and Rock Engineering, 2024, 57(2): 1353-1374.
[21]	余星辰,李小伟. 基于小波散射变换的煤矿瓦斯和煤尘爆炸声音识别方法[J]. 煤炭科学技术, 2024, 52(S1): 70-79.
	Yu X C, Li X W. Sound identification method of coal mine gas and coal dust explosion based on wavelet scattering transform[J]. Coal Science and Technology, 2024, 52(S1): 70-79.
[22]	Andreux M, Angles T, Exarchakis G, et al. Kymatio: Scattering transforms in python[A]. arXiv, 2018[2025-04-28]. .
[23]	Khodabandehlou H, Rashedi M, Wang T, et al. Cell culture product quality attribute prediction using convolutional neural networks and Raman spectroscopy[J]. Biotechnology and Bioengineering, 2024, 121(4): 1230-1242.
[24]	Li M-Y, Ebel B, Paris C, et al. Real-time monitoring of antibody glycosylation site occupancy by in situ Raman spectroscopy during bioreactor CHO cell cultures[J]. Biotechnology Progress, 2018, 34(2): 486-493.
[25]	Rashedi M, Khodabandehlou H, Wang T, et al. Integration of just-in-time learning with variational autoencoder for cell culture process monitoring based on Raman spectroscopy[J]. Biotechnology and Bioengineering, 2024, 121(7): 2205-2224.
[26]	Chen D, Chen Z W, Grant E R. Adaptive multiscale regression for reliable Raman quantitative analysis[J]. The Analyst, 2012, 137(1): 237-244.
[27]	Perezmarin D, Garridovaro A, Guerrero J. Non-linear regression methods in NIRS quantitative analysis[J]. Talanta, 2007, 72(1): 28-42.
[28]	Chen Z L, Jia K, Xiao C C, et al. Leaf area index estimation algorithm for GF-5 hyperspectral data based on different feature selection and machine learning methods[J]. Remote Sensing, 2020, 12(13): 2110.
[29]	Cui C H, Fearn T. Modern practical convolutional neural networks for multivariate regression: Applications to NIR calibration[J]. Chemometrics and Intelligent Laboratory Systems, 2018, 182: 9-20.
[30]	Zhang C, Springall J S, Wang X Y, et al. Rapid, quantitative determination of aggregation and particle formation for antibody drug conjugate therapeutics with label-free Raman spectroscopy[J]. Analytica Chimica Acta, 2019, 1081: 138-145.
[31]	Wang J R, Chen J Y, Studts J, et al. Simultaneous prediction of 16 quality attributes during protein A chromatography using machine learning based Raman spectroscopy models[J]. Biotechnology and Bioengineering, 2024, 121(5): 1729-1738.
[32]	Stone M T, Cotoni K A, Stoner J L. Cation exchange frontal chromatography for the removal of monoclonal antibody aggregates[J]. Journal of Chromatography A, 2019, 1599: 152-160.