Harnessing Artificial Neural Networks and large language models for bioprocess optimization: Predicting sugar output from Kraft waste-based lignocellulosic pretreatments

Abstract

This study implements Artificial Neural Network (ANN) models as predictive tools for glucose responses from Kraft waste-based pretreatments. The developed steam- and microwave-assisted ANN models achieved R2 scores > 0.95 for the observed and predicted glucose responses. An in-depth sensitivity analysis revealed that the glucose responses for the steam and microwave models were highly susceptible to the stepwise variation in green liquor dregs concentration (>3.3-fold) and power intensity (>2.6-fold), respectively. Comparative assessment on the capability of the large language model, ChatGPT, to generate innovative and factually accurate insights based on the process data was carried out. The novel process insights deduced by ChatGPT concurred with the authors’ findings of this study, underscoring the unique critical role of integrating advanced artificial intelligence and domain-specific knowledge to accelerate progression in lignocellulosic waste pretreatment. As such, these synergies align with global sustainable developmental objectives that leverage 4IR technologies, propelling this research field forward.