(149m) Hybrid Modeling to Simulate the Lignocellulosic Fermentation Processes Facing the Inhibitors’ Synergistic Effect and Mixed Carbon Sources

The urgent need to replace the current energy matrix of fossil fuels with renewable sources arises from the high economic reliance of several countries on oil, the volatility and instability associated with its price, and the environmental consequences that come with using fossil fuels. One example of this replacement is the use of lignocellulosic biomass to produce numerous value-added products including biofuels and chemicals. The conversion of lignocellulosic biomass into value-added product requires different processing techniques such as pre-treatment, enzymatic hydrolysis, and fermentation processes. The pre-treatment is an essential process to obtain fermentable sugars but producing several byproducts with inhibitory effects that synergistically interfere negatively in the fermentation. Such synergistic effect associated with the presence of mixed carbon sources, and the complex intracellular pathway of the microbial platform poses a challenge in the fully understanding and improvement of the lignocellulosic fermentation process (LFP). Integration of experimental observations and mathematical modeling is a powerful approach to analyze the fermentation comprehensively. Having a good model to predict process dynamic behavior is crucial for process operation and optimization, and to provide subsidies and tools for future modifications in the microbial platform structure [1,2].

Constructing first-principle model for LFP is challenging due to bioprocess high-level complexity and nonlinearity which cause mismatches between the real system observation and first-principle model prediction. On the other hand, data-driven model shows limited extrapolation ability which is harmful for process control and optimization. In turn, a hybrid model which aims to integrate mechanistic knowledge of process, and data-driven techniques to describe the unknown dynamics, offers a potential robust modeling strategy to model LFP, which is only partially understood, by exploiting the robust compartments of each strategy [3-7].

This work builds a hybrid model to simulate the complex dynamics of a LFP for process operation. The mass balances of each state variable are built up through the integration of mechanistic knowledge, related to Monod equation and glucose catabolic repression, and a data-driven technique which is an artificial neural network to represent the inhibitors synergistic effect that is not fully understood. Bayesian Information Criterion, a statistical model selection method that penalize the goodness of fit between the model prediction and the experimental data in terms of number of parameters, is used to determine the best neural network structure [8]. The hybrid model is first constructed to model the acetic acid inhibition effect. Then, the hybrid model is further improved by accounting the impact of up to seven six inhibitors’ compounds based on data from Li et al., 2017 [9]. Global Sensitivity Analysis is also performed to quantify the importance of model inputs and their interactions with respect to model outputs. The results are very promising for digital simulation and operation of the LFP, and to account the inhibitors’ synergistic effect. A schematic representation of the current work is shown in Figure 1.

References

[1] CABANEROS LOPEZ, P. et al. Transforming data to information: A parallel hybrid model for real‐time state estimation in lignocellulosic ethanol fermentation. Biotechnology and bioengineering, v. 118, n. 2, p. 579-591, 2021.

[2] LEE, D.; JAYARAMAN, A.; KWON, J. S. Development of a hybrid model for a partially known intracellular signaling pathway through correction term estimation and neural network modeling. PLoS Computational Biology, v. 16, n. 12, p. e1008472, 2020.

[3] NARAYANAN, H. et al. Bioprocessing in the digital age: the role of process models. Biotechnology Journal, v. 15, n. 1, p. 1900172, 2020.

[4] TSOPANOGLOU, A.; DEL VAL, I. J. Moving towards an era of hybrid modelling: advantages and challenges of coupling mechanistic and data-driven models for upstream pharmaceutical bioprocesses. Current Opinion in Chemical Engineering, v. 32, p. 100691, 2021.

[5] ZHANG, D.; SAVAGE, T. R.; CHO, B. A. Combining model structure identification and hybrid modelling for photo‐production process predictive simulation and optimisation. Biotechnology and Bioengineering, v. 117, n. 11, p. 3356-3367, 2020.

[6] PRADHAN, D.; JAISWAL, S.; JAISWAL, A. K. Artificial neural networks in valorization process modeling of lignocellulosic biomass. Biofuels, Bioproducts and Biorefining, v. 16, n. 6, p. 1849-1868, 2022.

[7] WANG, H.; KONTORAVDI,C.; DEL RIO CHANONA, E. A. A Hybrid Modelling Framework for Dynamic Modelling of Bioprocess. 33rd European Symposium on Computer Aided Process Engineering (ESCAPE33), Athens, Greece, 2023 – Accepted Paper.

[8] DING, J.; TAROKH, V.; YANG, Y. Model selection techniques: An overview. IEEE Signal Processing Magazine, v. 35, n. 6, p. 16-34, 2018.

[9] LI, Y. C.; GOU, Z. X.; ZHANG, Y.; XIA, Z. Y.; TANG, Y. Q.; KIDA, K. Inhibitor tolerance of a recombinant flocculating industrial Saccharomyces cerevisiae strain during glucose and xylose co-fermentation. Brazilian Journal of Microbiology, v. 48, p. 791-800, 2017.