Raman+NIR Spectra Improves Glucose Soft Sensing
Combining Raman and NIR spectroscopy into a single framework creates a multi-modal soft-sensing option that is effective for fed-batch, online, and in-line glucose sensing, preparing the way for hybrid/continuous processing. The post Raman+NIR Spectra Improves Glucose Soft Sensing appeared first on GEN - Genetic Engineering and Biotechnology News .
In biopharmaceutical manufacturing, precise glucose monitoring is crucial, particularly as companies adopt hybrid and continuous bioprocessing for advanced therapies. Researchers have developed a novel multimodal soft-sensing framework that combines Raman and near-infrared (NIR) spectroscopy to overcome the limitations of traditional sensors. This innovative approach offers real-time, high-frequency, and low-latency measurements, which are not typically available with existing commercial offline sensors.
The framework, developed by a team including Xiwei Tian and Yinlan Ruan, encodes each spectroscopy modality separately before merging them using a dual-stream, cross-attention transformer architecture (DS-CAT). When tested on 142 preprocessed, synchronized Raman-NIR records, the model explained 99% of the variation in true glucose concentration, meeting the good performance threshold of above 95%.
The root mean square error of 1.09 g/L and mean absolute error of 0.89 g/L further demonstrate its accuracy and effectiveness in ensuring proper feed control.
Further analysis revealed the strong dependence on Raman information, with the interval 1122−1335 cm⁻¹ (associated with carbohydrate vibrational modes) identified as the most sensitive. These findings, when compared to other literature reports, demonstrate the enhanced sensitivity, reduced noise, and reduced drift of the DS-CAT framework.
The results also indicate that this approach outperforms traditional modeling methods such as partial least squares regression, gradient boosting regression trees, convolutional neural networks, and temporal convolutional networks.
By delivering high-frequency and low-latency data, the DS-CAT approach has the potential to significantly improve soft-sensor capabilities for real-time glucose monitoring. This could lead to better control of a critical parameter in closed-loop mammalian and microbial fermentation, ultimately enabling the rational transition from fed-batch fermentation toward continuous manufacturing.
Written by urgent.news from GEN Biotechnology's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
