AI-generated from publicly available materials.Recent advancements in glucose monitoring techniques could significantly enhance biopharmaceutical manufacturing processes. A new multimodal soft-sensing framework that combines Raman and near-infrared (NIR) spectroscopy has emerged, addressing long-standing challenges in accurate glucose measurement during fed-batch fermentation.
This innovative framework enables high-frequency and low-latency glucose measurements, which are often lacking in traditional offline sensors. The approach leverages the distinct advantages of both Raman spectroscopy, known for its precise chemical fingerprints, and NIR spectroscopy, which provides extensive spectral information. Researchers from East China University of Science and Technology and Guilin University of Electronic Technology developed this system using a dual-stream, cross-attention transformer architecture (DS-CAT) to effectively integrate the two modalities.
Testing revealed that the DS-CAT model accounted for 99% of the variance in glucose concentrations, surpassing the 95% threshold typically deemed acceptable. With a root mean square error of 1.09 g/L and a mean absolute error of 0.89 g/L, the model demonstrates the reliability necessary for effective feed control in real-time processing. Additionally, sensitivity analyses indicated that the Raman component significantly influenced prediction accuracy, particularly in specific vibrational intervals associated with carbohydrates.
Crucially, the DS-CAT framework outperformed established modeling techniques such as partial least squares regression and convolutional neural networks. This positions it as a promising solution for enhancing soft-sensor capabilities in glucose monitoring, potentially facilitating a shift from fed-batch to continuous bioprocessing. The implications of these findings could lead to more efficient control of fermentation processes, ultimately improving the production of biopharmaceuticals.