Optimizing Saccharomyces cerevisiae biomass production with Ziziphus lotus juice via hybrid modeling approaches

The development of sustainable and cost-effective fermentation processes requires the use of unconventional substrates and optimization strategies. In this study, Ziziphus lotus juice (ZLJ) was investigated as a novel carbon source for Saccharomyces cerevisiae biomass production. A Plackett-Burman design was applied to identify critical factors, with sugar concentration, yeast extract (YE), MgSO₄, and MnSO₄ emerging as the most influential. Response surface methodology (RSM) was then combined with artificial neural networks (ANNs) to capture nonlinear factor interactions. The optimal ANN architecture achieved high predictive accuracy, and its applicability domain confirmed robust generalization across training, validation, and testing datasets. To further refine medium composition, a genetic algorithm (GA) was integrated with the RSM-ANN model. The optimized conditions (5.27 g/L YE, 0.01 g/L MgSO₄, 0.05 g/L MnSO₄, 80 g/L ZLJ) yielded a biomass production of 7.23 g/L, representing a twofold increase compared to the initial medium (3.81 g/L). The RSM-ANN-GA framework outperformed RSM, reducing prediction errors and enhancing model reliability. Results highlight the potential of ZLJ as a cost-effective, nutrient-rich substrate and demonstrates the power of hybrid computational modeling for fermentation optimization. The integrated RSM-ANN-GA approach offers a scalable and reproducible strategy for improving yeast biomass production, with broader applicability to microbial bioprocesses and industrial biotechnology. This work is the first to combine Z. lotus juice valorization with a hybrid RSM-ANN-GA framework for yeast biomass optimization, enabling efficient modeling of nonlinear nutrient interactions. The proposed strategy demonstrates clear advantages over conventional RSM and provides a transferable approach for optimizing fermentation processes using unconventional substrates.