Reliable calibration and validation of phenomenological and hybrid models of high-cell-density fed-batch cultures subject to metabolic overflow

• Published in: Computers & chemical engineering Vol. 186; p. 108706
• Main Authors: Ibáñez, Francisco, Puentes-Cantor, Hernán, Bárzaga-Martell, Lisbel, Saa, Pedro A., Agosin, Eduardo, Pérez-Correa, José Ricardo
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2024
• Subjects: Dynamic optimization; Fed-batch fermentation; High-density cultures; Hybrid models; Overflow metabolism; Physics-informed neural networks; Overflow metabolism; High-density cultures; Hybrid models; Physics-informed neural networks; Dynamic optimization; Fed-batch fermentation
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2024.108706

Fed-batch cultures are the preferred operation mode for industrial bioprocesses requiring high cellular densities. Avoids accumulation of major fermentation by-products due to metabolic overflow, increasing process productivity. Reproducible operation at high cell densities is challenging (>100 gDCW/L), which has precluded rigorous model evaluation. Here, we evaluated three phenomenological models and proposed a novel hybrid model including a neural network. For this task, we generated highly reproducible fed-batch datasets of a recombinant yeast growing under oxidative, oxygen-limited, and respiro-fermentative metabolic regimes. The models were reliably calibrated using a systematic workflow based on pre-and post-regression diagnostics. Compared to the best-performing phenomenological model, the hybrid model substantially improved performance by 3.6- and 1.7-fold in the training and test data, respectively. This study illustrates how hybrid modeling approaches can advance our description of complex bioprocesses that could support more efficient operation strategies. •High-cell density fed-batch cultures were performed under different regimes.•Phenomenological models were calibrated and evaluated in different conditions.•Best performing model was expanded with a neural network model.•Hybrid model displayed superior accuracy for fitting training data and predicting test data.