Bayesian inference on fermentation kinetics: Comparative analysis with frequentist approach

• Construct a Bayesian hierarchical regression model with modified hyperbolic secant function. • Compare Bayesian approach to frequentist counterpart on fermentation data. • Priors ensure biological plausibility where frequentist estimates fail. • Bayesian inference resolves convergence issues in sparse data scenarios. This study explores the application of Bayesian inference to fermentation kinetic modeling, focusing on two representative batch systems: glycerol-glucose co-fermentation for 1,3-propanediol (1,3-PDO) production and dark fermentation for biohydrogen production. Modified hyperbolic secant functions were implemented to model metabolite concentrations obtained from above-noted experiments. Then, the performance of a Bayesian hierarchical model to that of frequentist nonlinear least squares fitting was compared. The results indicate that while the frequentist approach demonstrates high computational efficiency and suitability for short-term prediction, the Bayesian approach offers significant advantages in terms of parameter interpretability, robustness under limited data, and uncertainty quantification. Overall, Bayesian methods enhance the construction of interpretable, stable, and data-efficient fermentation kinetic models.