Accelerating process development of recombinant protein in Escherichia coli fermentation through deep learning and Raman process analysis techniques
摘要
In the context of biopharmaceutical production, improving the yield of recombinant proteins is essential for industrial-scale processes. This study developed a genetic algorithm-based semi-supervised convolutional neural network (GA-SCNN) to convert Raman spectra into real-time predictions of target protein (ProA5m) expression, enabling accelerated optimization of fermentation. Using a defined culture medium to minimize spectral interference, the GA-SCNN model was trained on fermentation data and showed accurate real-time monitoring of protein levels in new 5 L fermentations. Separately, Design of Experiments identified cooling temperature after induction, feed rate, and induction time as key factors influencing yield. Under optimized conditions, namely a cooling temperature of 22 °C, an induction time of 10.3 h, and a feed rate of 0.27 h− 1, the target protein yield increased by 335%, rising from 2.80 g L− 1 to 9.37 g L− 1. The GA-SCNN model achieved acceptable predictive accuracy and, together with its visualization capability, contributed to shortening the fermentation cycle, reducing contamination risks, and improving development efficiency. This work successfully integrates Raman spectroscopy with artificial intelligence, providing substantial value for biopharmaceutical process development.