Ethanol Production From Pretreated Potato Residues By Co-fermentation of Saccharomyces cerevisiae and Pichia stipitis: Harnessing Large Language Models for Process Optimization

Abstract

The need for alternative, renewable, and sustainable energy sources, driven by climate change and the detrimental effects of fossil fuel consumption, necessitates innovative solutions. Thus, this study investigates the optimization of ethanol yields from pretreated potato residues, using a co-culture of Saccharomyces cerevisiae and Pichia stipitis. Response Surface Methodology (RSM) was initially used to model the co-fermentation process, examining solid loading (5–20%), inoculation ratio (1—4), and inoculation time (0—18 h) as key factors. Thereafter, the study evaluated the performance of large language model (LLM), GPT4o mini, in generating insightful co-fermentation optimization process data. The RSM model demonstrated a high coefficient of determination (R2 = 0.97) and a significant p-value (0.0016). Optimal conditions identified were 12.5 w/v solid loading, an inoculation ratio of 4:1 (4 S. cerevisiae: 1 P. stipitis), and an inoculation time of 0 h, with ethanol yield of 0.544 g/g-utilized sugar. The LLM predicted two different (6 and 12 h) inoculating times that was combined with the same 12.5 w/v solid loading, and the inoculation ratio of 4:1. Experimental RSM validation gave ethanol yields of 0.501 g/g-utilized sugar for the co-culture, 0.485 g/g-utilized sugar for S. cerevisiae mono-fermentation, and 0.076 g/g-utilized sugar for P. stipitis mono-fermentation, indicating a 1.03-fold and 6.59-fold enhancement in ethanol yield in the co-culture system compared to the mono-fermentations, respectively. While the experimental LLM validation showed ethanol yields of 0.504 g/g-utilized sugar and 0.501 g/g-utilized sugar for the 6 h and 12 h respectively. The synergistic effects of S. cerevisiae and P. stipitis in this optimized co-culture system, along with the integration of RSM and LLM approaches, demonstrate the potential of LLM for enhancing co-culture ethanol production from lignocellulosic biomass in the nearest future. 1st report on response surface methodology & ChatGPT for co-fermentation optimization. RSM-optimized conditions: 12.5% solid loading, 4:1 inocula ratio, and 0-h inoculation time. Co-culture achieved ethanol yield of 0.501 g/g, outperforming mono-fermentation yields. Insights from Large Language Models (LLMs), using retrieval-augmented generation (RAG). LLMs staggered inoculation strategies ethanol yields were comparable to RSM-optimized yield. Early co-inoculation enhanced sugar metabolism towards higher ethanol yields.