Department of Systems Biology, Technical University of Denmark, Kongens Lyngby, Denmark.
J Biotechnol. 2009 Oct 26;144(2):102-12. doi: 10.1016/j.jbiotec.2009.08.018. Epub 2009 Sep 6.
Batch bioreactor cultivations using Saccharomyces cerevisiae at high (190-305 gl(-1) glucose) or low (21-25 gl(-1) glucose) gravity conditions were monitored on-line using multi-wavelength fluorescence (MWF) and standard monitoring sensors. Partial least squares models were calibrated for the prediction of cell dry weight (CDW), ethanol and consumed glucose, using the two data types separately. The low gravity cultivations (LGCs) consisted of two phases (glucose consumption with concomitant ethanol production followed by ethanol consumption after glucose depletion), which proved difficult to model using one and the same model for both phases. Segmented modelling, using different models for the two phases, improved the predictions significantly. The prediction models calibrated on standard on-line process data displayed similar or lower root mean square error of prediction (RMSEP) compared to the fluorescence models. The best prediction models for high gravity cultivations (HGCs) had RMSEPs of 1.0 gl(-1) CDW, 1.8 gl(-1) ethanol and 5.0 gl(-1) consumed glucose, corresponding to 4%, 2% and 2% of the respective concentration intervals. Corresponding numbers in low gravity models were 0.3 gl(-1) CDW, 0.7 gl(-1) ethanol and 1.0 gl(-1) consumed glucose, corresponding to 4%, 8% and 4% of the respective concentration intervals.
采用多波长荧光（MWF）和标准监测传感器，对酿酒酵母在高（190-305 g/L 葡萄糖）或低（21-25 g/L 葡萄糖）重力条件下的批式生物反应器培养进行在线监测。使用两种数据类型分别为细胞干重（CDW）、乙醇和消耗葡萄糖的预测建立偏最小二乘模型。低重力培养（LGC）由两个阶段组成（葡萄糖消耗伴随着乙醇生产，然后在葡萄糖耗尽后消耗乙醇），这两个阶段很难用同一个模型进行建模。分段建模，为两个阶段使用不同的模型，显著提高了预测能力。基于标准在线过程数据校准的预测模型与荧光模型相比，预测值的均方根误差（RMSEP）相似或更低。高重力培养（HGC）的最佳预测模型的 RMSEP 分别为 1.0 g/L CDW、1.8 g/L 乙醇和 5.0 g/L 消耗葡萄糖，分别对应于相应浓度范围的 4%、2%和 2%。低重力模型的相应数值分别为 0.3 g/L CDW、0.7 g/L 乙醇和 1.0 g/L 消耗葡萄糖，分别对应于相应浓度范围的 4%、8%和 4%。
